Latest Posts (20 found)

Obituaries

Listen to this post: Dearly beloved, we are gathered here today to get through this thing called life. ―  Prince Jensen Huang resigned as CEO of Nvidia on Thursday, November 12, 2026. He had been CEO of the company ever since he founded it on April 5, 1993. Nvidia started out as a graphics card company; Huang’s exit — nominally because of affairs with a buyer who may or may not have been a Chinese spy, as well as at least two company employees — was ultimately because Nvidia was, by the powers that be, considered a defense contractor, and Huang never adjusted. In this, Huang was the ultimate victim of his own success. Nvidia’s first graphics cards simply translated triangles into pixels. The 2001 GeForce 3 introduced a programmable pixel shader, which meant a video card’s functionality was no longer pre-defined. The killer move, however — which almost killed the company — was the 2006 introduction of CUDA, which was an API for programming those shaders. For much of the next 16 years Huang sold the gospel of CUDA , convinced it was the platform for a future of accelerated computing; then, in 2022, came ChatGPT. ChatGPT made clear what GPUs were most useful for: large language models, and Nvidia made the best GPUs. More than that, Huang had realized that the future of accelerated computing was not just chips, but rather systems; that drove the purchase of Mellanox in 2019, which provided the technology for Nvidia to scale out from chips, setting it down the road to being the best option for the sort of large scale training that made large language models possible. For this Nvidia charged very high prices, with margins exceeding 70%, because they could: they had the platform, the technology, and, over the past few years as others sought to make their own chips, the supply chain. Huang’s impact on the Asian supply chain in particular cannot be overstated. Byrne Hobart, in his book Boom: Bubbles and the End of Stagnation , notes that one of the benefits of bubbles is that they provide a coordination function, as the fear of missing out incentivizes multiple entities with interlocking dependencies to all commit to building at the same time; Huang was, in many respects, a one-person coordination function. Huang gave multiple keynotes a year, which he explained as how he communicated with the ecosystem broadly, and was famous for eating and drinking his way across Asia in particular, convincing reluctant suppliers that the AI boom was real, and locking up the lion’s share of their output for Nvidia and its stunning commitment to a yearly release cycle. Huang, who had toiled for decades in relative obscurity, known primarily to gamers, and only barely to the tech industry, became a rock star, swarmed wherever he went, always in his black leather jacket. He also, to his ultimate detriment, became a political figure. Nvidia had long had no presence in either Washington D.C. or Beijing , and was blindsided by the Biden administration’s imposition of chip controls in 2022. Nvidia’s immediate reaction to those initial controls, which restricted exports based on interconnect speed, was to immediately release the A800 in China, which was an A100 with a third of its interconnects disabled. That set the tone for the conflict that I believe was the true cause of Huang’s downfall: Huang strenuously disagreed with the various iterations of the chip ban, and Nvidia only bothered to do the bare minimum to abide by the letter of the law, never the spirit. That approach, in all likelihood, extended to Nvidia’s oversight (or lack thereof) of Singaporean entities that were alleged to be fronts for chips for China. Some of these were legal: Chinese companies are allowed to access cloud computing facilities outside of China that house leading edge Nvidia chips, although chip control advocates are seeking to close that loophole; some were not legal, in that the chips were ultimately ferried to China. Megaspeed, whose executive, Alice Huang, was the first person Huang was alleged to have an affair with, was officially in the former business, and alleged to actually be in the latter. There are whispers that Megaspeed’s ultimate customer was not, or not just, Alibaba, but also the Chinese military. These are whispers, to be clear; it was the affairs that were enough, at least publicly, to lead to Huang’s resignation. Nvidia’s durability — at least that of their margins — was already an open question. Start with CUDA: when the market was software written for GPUs, a platform — including not just an API but a huge set of libraries targeting every conceivable market for accelerated computing — was a nearly impenetrable moat. That is why LLMs like GPT were written using CUDA. LLMs, though, both brought Nvidia its incredible success, and also seeded its greatest risks. The first is that the vast majority of developer energy moved up the stack, using the LLM’s API as a platform; for most companies, CUDA didn’t matter. The second is that Anthropic, which found funding from Amazon and Google, originally also-rans to the Microsoft-OpenAI partnership, had to learn how to use Trainium chips from Amazon and TPUs from Google; this was difficult, but ultimately meant that the leading LLM did not have any Nvidia dependency at all (Anthropic runs on Nvidia chips, because they learned how to run on anything). The third is the most devastating: LLMs are so effective at writing software that any software-based moat will inevitably dry up. Nvidia’s second moat, its networking and systems approach, is more durable, although challenged by Google (and by extension, Anthropic) in particular. Nvidia’s systems approach is also the easy answer for enterprises and governments seeking to own their own hardware, and neoclouds seeking to build data centers that are independent of the hyperscalers. The latter market rents chips to both hyperscalers and frontier labs, as well as individual companies, but is challenged as obvious sources of power and political permission disappear. They also need money to afford Nvidia’s systems, which has increasingly come from Nvidia itself. Nvidia’s third moat is its mastery of the supply chain, and this is the moat that suddenly looks much more tenuous. Huang has bragged about the fact that he has never signed a contract with TSMC over their three decades of partnership; it’s a relationship built on a handshake and mutual trust with Huang specifically. This is, per the point above, a very Asian way of doing business, and Huang was the master of that. It’s questionable whether or not that trust and willingness to commit will persist when the handshake is no longer there. Nvidia’s most important moat, however, was the degree to which it was not a normal company. Huang organized Nvidia as an exoskeleton around one man: himself. There is very little management at Nvidia — Huang famously has 60 direct reports — for the same reason that there was no succession plan: Nvidia could move so quickly and so aggressively because the company still behaved as if it were a startup. Indeed, lots of established companies — including those still led by founders — claim to still be a startup, but only one, Nvidia, really was organized that way. That paid off with all of those successful bets; it also created an environment where there were no checks on Huang. And, now that he is gone, the exoskeleton is much more poised to collapse than you might expect. The first part of the AI buildout was funded by free cash flow, particularly from the hyperscalers. Then, during the second half of last year, nearly everyone started turning to debt. Now — starting with Google this past summer — the primary funding mechanism is equity. In the long run, this should come full circle, and future build-outs will once again be funded by free cash flow, generated from the infrastructure that is being built now. The problem for Nvidia is that the companies most able to navigate this progression are the hyperscalers, which have the resources to challenge Nvidia’s hold on the chip market. To that end, Nvidia has been very eager to ensure that it is selling not only to the hyperscalers, but to other would-be competitors, like the neoclouds; there will likely, at some point, be serious questions raised about exactly how Nvidia went about allocating compute in pursuit of this goal. The neoclouds, however, are at a meaningful disadvantage when it comes to funding. First, their debt is significantly more expensive, given that their only business is renting GPUs (as opposed to, say, running the most valuable advertising businesses in the world, or the world’s largest e-commerce platform, or dominating enterprise software). Second, their equity is increasingly worthless, given the significant risks attached to that credit. If a neocloud can’t pay, it’s the equity holders that get wiped out first. This led to what I termed Nvidia’s Risky Business : Nvidia increasingly used its cash flow and credit rating to backstop and fund the companies who were buying its chips. Bears called this circular financing; bulls called it a necessary bridge to a future where AI revenue paid down these companies’ debt and drove massive upside. What is undoubtedly true is that this was the sort of risky business that is uniquely possible for a multi-trillion dollar company that is completely controlled by one man, who has made an entire career of pushing the limits and making risky bets — and seeing them pay off. It’s also a series of bets that are unlikely to continue. The neoclouds should be very nervous about losing their patron, along with their creditors and equity holders. The other problem is that the revenue necessary to make all of this build-out work is something on the order of a trillion dollars industry-wide, and it needs to materialize sooner rather than later; if it falls short, it’s the hyperscalers who will prioritize their own compute first, and they will likely be seen as a safe haven for everyone committing to spending. In that world, Nvidia’s neocloud hedge — absent Huang’s willingness to shore the ecosystem up — is liable to turn into a millstone, with their assets flooding the market even as the supply chain ramp-up brings ever more supply to market. Perhaps because he had been preaching the gospel of accelerated computing for so long, Huang was the leading figure pushing back against the doomer narrative surrounding AI. Yes, AI was a massive business opportunity, but it was still normal technology; the safety issues presented were engineering challenges to be solved. What required consideration was not just the next five years, but the next fifty. It’s a view that I am, in full disclosure, generally aligned with. Many others are not. On one hand there are the aforementioned doomers, certain that fast takeoff is imminent — or has already started — as AI increasingly builds better AI in a recursive loop. The end state is an AI that is massively more capable than humans, and uncontrollable by them; the only hope is to ensure that AI is aligned with humanity, by which adherents to this viewpoint mean aligned with them and their view of the world. This contingent is, broadly speaking, represented by Anthropic. And, in this view, actively working to enable and accelerate uncontrollable open weights models is a threat to humanity. On the other hand there are the China hawks, who view AI as a national security issue, and the key to U.S. military dominance in a world where China owns the means of production of most militarily relevant hardware. In this world giving China access to chips is effectively treason against the United States. For both of these groups, Huang was public enemy number one, because of his determination to do what was best for Nvidia’s business, which was to sell chips to China. This made sense for Nvidia for multiple reasons: This last point is worth dwelling on. The chief objection that China hawks in particular make to the idea that it is a good idea to sell China chips in order to prevent them from surpassing the U.S. is that China is already heavily invested in building their own ecosystem. This is certainly true, but I think it underplays the complexity of chips and what is necessary to actually compete with the West. Specifically, the Chinese government can only control the supply side, by spending money and making top-down imperatives; building an ecosystem, however, whether that be for AI specifically, or semiconductors broadly, requires demand as well. Progress in models and systems and chips depends on a feedback loop between the user and the developer, or the customer and the manufacturer. The demand side, however, is much more short-term oriented than the Chinese government might prefer. The model companies’ concern is the other model companies: that means if they are distilling, we must distill too. If they are using Nvidia chips, then we must too. This matters up and down the stack, just as it has always mattered in technology, which really is different in this regard from almost every other industry. A useful analogy is to consider the Internet. Why is it that country after country can profess a desire to have an indigenous alternative to the American tech giants, yet miserably fail in every attempt to generate it? It’s because the tech giants are only a click away! In this regard, China’s Great Firewall was one of the most brilliant geostrategic moves ever: by virtue of cutting off access to the American tech companies, China created the demand that pulled China’s own tech giants into existence. You need demand, not just desire. This is why China itself is currently banning the import of Nvidia’s two-generations-behind H20 chips, after the Trump administration foolishly banned their import, only to reverse themselves while bragging about the reason to do so — i.e. keep China addicted to Nvidia. China is replaying the Great Firewall playbook: force its internal demand to use its much worse alternatives, because that is the only way those alternatives become better. And, in the long run, those alternatives, and the entire ecosystem that supports them, will flood the world. Huang was, to be perfectly clear, talking his book when it came to his insistence on selling chips to China, or, in the case of the current controversy, allegedly turning a blind eye towards nominally legal buyers who were intent on violating the chip ban; that doesn’t mean, however, that he was wrong about the long-term effects, at least if AI is a normal technology (and I haven’t gotten to the risks this imposes on the Taiwan Strait ). What is clear now, however, is that the doomers and the hawks have won: AI, at least in terms of policy, will be treated as anything but normal, and in the fullness of time, it is the U.S. chip ecosystem that will pay the ultimate price. The birth date for the last obituary was the invention of the transistor; the birth date for this one is the founding of Intel. Robert Noyce and co-founders had spent the last decade at Fairchild Semiconductor, but were constrained by their parent company’s fundamental conservatism and chafed at their lack of upside. Arthur Rock effectively created the venture capital industry in Silicon Valley through his financing of Intel; Noyce created the concept of the commercial-market-led technology company that made it up in volume. At that time the largest single purchaser of chips was the U.S. government; Noyce realized that the fastest way to progress was to make chips for everyone, including the government, and on his terms, not theirs. From that point on, tech as an industry operated with the assumption that the world was its customer, and that their own executives defined the route to progress. Government was free to buy their products, but not to define them, and if they wanted the best products, then they best not restrict who they could sell them to either. That’s how important volume was, and, over time, it became clear how powerful and essential ecosystems were. Other industries operate differently; I thought this article in the Financial Times was striking, and not just because of the headline: Bessent says Nvidia needs ‘shake-up’ with external CEO : Bessent said that it is important “to examine whether Nvidia had a culture of compliance around the US export restrictions” and whether its top executives knew they were being skirted. “Was it a low-level fail? Was it in the middle, or was it up in the C-suite? And who knew? These shell companies were obvious, they were being investigated, and did Nvidia use them as a workaround to sell chips to the Chinese?” he said. “Were they pressing these seemingly illicit sales to goose their revenues and their earnings?” The Treasury secretary added that Nvidia needs to provide answers within a matter of days: “I think we need answers in the next 10 days from the board and from the outside counsel”. He added that “we will see if Nvidia intentionally ignored the warning signs of these shell companies, many of them in Singapore”. Bessent has pointed the spotlight on Howard Lutnick’s commerce department for failing to ban exports of Nvidia’s chips to Megaspeed, the Singaporean company linked to the alleged Chinese spy who had an affair with Huang. US authorities have been investigating whether Megaspeed, which builds AI data centres in Asia, has been involved in smuggling high-end Nvidia chips, which are powering the AI boom, into China. Bessent has blamed the commerce department more than once this week for failing to ban exports of Nvidia’s chips to Megaspeed. Commerce delayed efforts to add Megaspeed to Washington’s blacklist of companies the US considers a risk to national security. “Many other departments were pushing for this, and commerce had the pen,” Bessent repeated on Sunday. “We have all the rules we need. We just need them enforced.” Lutnick is under mounting pressure from within the administration and from Congress, which has demanded that he explain how he allowed chips to be exported to a company linked to an alleged Chinese spy. US lawmakers are also accelerating efforts to crack down on advanced AI chips reaching China. If Nvidia “were a bank and they hadn’t complied like this, the [Treasury’s Office of the Comptroller of the Currency] would be in there right now, and a lot of the executives would be removed”, Bessent said. Scott Bessent is the Treasury Secretary, which is to say that he operates an apparatus that is arguably just as, if not more, powerful than the U.S. military. Treasury controls the flow of money around the world, and its soldiers are the banks; if Treasury has sanctioned an entity, then it is up to banks to ensure those sanctions are enforced. They need to know their customers, and they need to analyze every dollar flowing through their systems; if they do not, they are breaking the law. Ignorance is not a defense. Ignorance, on the other hand, has been tech’s defense basically forever. Yes, there are bad people in the world, which means that if the whole world is using your products, they will be used to do bad things. That was considered an acceptable trade-off, however, because most people are good and the benefits that come from scale — and an avoidance of the friction that comes from trying to track the activities of billions of people — were worth the risk. This was certainly what Huang believed, and, to his detriment, he aggressively belittled anyone who dared disagree (the extent to which Huang is despised in Washington D.C., as evidenced by the giddiness and bloodlust online last week, is extraordinary). In Huang’s view, as long as the Commerce Department said that it was ok to sell to Megaspeed, then it was ok to sell to Megaspeed — and to sleep with one of its executives. Bessent views the matter entirely differently: he thinks that Nvidia should operate like a bank. It’s not just their job to abide by the letter of the law, but the spirit of it. It’s not the government’s job to enforce the chip ban, but the company’s. An attestation from a Singaporean customer that chips aren’t going to China isn’t worth anything; actually putting boots on the ground and ensuring compliance is the only way Nvidia cannot be in violation of the U.S. government’s dictates. Again, for tech, this is extraordinary, even if it is normal in banking. Extraordinary enough, in fact, that Huang is being labeled by many — not just in Washington, but also in the finance world — as a traitor. Huang, in my estimation, is just a businessman who cares about nothing else other than Nvidia; that worked out great when he made GPUs for gaming that no one in government cared about or understood. That’s just a normal technology; when AI was deemed to be anything but normal, however, business as usual wasn’t just over: it was a crime, and it cost Huang his job — and maybe more. It’s worth pointing out that while this episode was shocking, I should not, in retrospect, have been surprised. Earlier this year, when Anthropic faced off against the U.S. government, I wrote in Anthropic and Alignment : At the same time, what is the standard by which it should be decided what is allowed and not allowed if not laws, which are passed by an elected Congress? Anthropic’s position is that Amodei — who I am using as a stand-in for Anthropic’s management and its board — ought to decide what its models are used for, despite the fact that Amodei is not elected and not accountable to the public. And, on the second point, who decides when and in what way American military capabilities are used? That is the responsibility of the Department of War, which ultimately answers to the President, who also is elected. Once again, however, Anthropic’s position is that an unaccountable Amodei can unilaterally restrict what its models are used for. My point in that Article was that if AI is as exceptional as claimed, then the government will demand a say in how it is used, whether or not private company executives agree; moreover, that was a good thing, or at least a least-bad thing. The same analysis applies to Nvidia and Huang: Huang may have fundamentally disagreed with the chip ban, but at the end of the day, it was his duty to accept that it was the law of the land. To the extent he did not do so is ultimately his responsibility. Perhaps the most tragic aspect of the Huang affair is the extent to which he was the personification of The American Dream. Huang was sent to America as a nine year old who didn’t speak a lick of English, and somehow ended up at a boarding school in rural Kentucky for misbehaving youth, where he would trade help on homework for protection from bullies. From there he went to a state school in Oregon, became an engineer, before famously ending up at a Denny’s diner where he and his co-founders decided to start Nvidia. Then, over the following 33 years, he took risk after risk, created innovation after innovation, until he became a hero not just in the United States, but across the world. And then, you might say, he flew too close to the sun. Certainly he did so in his private life: it’s not just the affairs, which may be par for the course for rich and powerful men; rather, it’s notable just how reckless the affairs were. The ones that were revealed last week were with a customer and potential Chinese spy, and two of his own employees; both are fireable offenses. Meanwhile, the way in which Huang treated his opponents — including actively seeking to get people fired who he felt stood in the way of getting his way — bore the marks of a man who has constructed a life, and a company, where no one can or will stand up to him. That is a recipe for disaster, and here we are. At the same time, it really does bum me out that what ultimately doomed Huang was a set of rules that really are a big departure from the status quo — and, in my opinion, misguided. Huang’s actions — or lack of actions, in terms of enforcing compliance with the chip ban — are treated as if they are an obvious crime, when in fact this is a dramatic shift in expectations from a short time ago (although there is certainly a callback to previous controversies around selling advanced computing to adversaries). Huang should have adjusted and accepted the new reality, make no mistake; I’m sympathetic to the fact he couldn’t, and recognize the failure in not having structures in place that could have warned him about the dangerous road he was embarking on, even as I realize Huang’s willingness to navigate dangerous roads is precisely why he was an American success story. Beyond Huang, however, there is the question of how we compete with China, and I am disappointed, although not surprised, that we are pursuing the route of bans and forced compliance, instead of innovation and openness. Make no mistake, I am no China dove, and I have been very early to the argument that we erred greatly in willingly exporting entire industries to the communist state. That, however, makes it even more imperative that we sprint in the opposite direction of China, towards more freedom in a bet on inventing new industries, not more control and a focus on gatekeeping what we have. I get the logic, particularly in the case of AI, which is extraordinary; I lament the implications of what it says about our belief in ourselves and our ability to stay ahead. Presented by As AI models improve, you might expect the market will consolidate around a handful of providers. But user data tells a different story. Among OpenRouter customers spending at least $100 a month, the median number of AI model families in use has doubled over the past year , from three to six. Even as AI models become more capable, businesses are using more models, not fewer. This makes sense when you consider that intelligence isn’t one-dimensional. A model that’s exceptional at writing code may not be the fastest at answering customer questions, or the most economical at processing millions of documents. The question is shifting from “which AI model should we build on?” to “which model should handle each task?” The answer, increasingly, is ‘it depends.’ Which is great for flexibility, less great for simplicity. OpenRouter, which recently joined Stripe , allows businesses to access and govern hundreds of models with the flexibility to optimize for cost, performance, and reliability as the market evolves. Learn more at openrouter.ai . The first and most obvious reason is that it is good for a company to be able to sell to everyone, including the second largest market in the world (China has traditionally made up about a quarter of Nvidia’s sales, long before AI was even a thing). The second is that China is a large enough market, with sufficiently motivated and capable developers, that it can actually create a meaningful CUDA alternative; furthermore, such an alternative would, in the long run, not be constrained to China, but rather compete with Nvidia all over the world. The third is that China’s AI ecosystem, unconstrained as it is by such niceties as terms of service, distills frontier lab models and releases them as open weights models; those models, meanwhile, are tuned to run on Nvidia chips. These models give Nvidia leverage versus the frontier labs all over the world, including in the West. The fourth is that China is uniquely positioned to create a fundamentally different approach to systems design using optical interconnects to scale out training clusters. This is much more inefficient in terms of power usage than Nvidia’s more elegant designs, but China is also much more capable of manufacturing excess amounts of power to make it work. The fifth is that, in the very long run, China can create an entire alternative supply chain for semiconductors broadly, rendering Nvidia’s mastery of the Western-aligned supply chain much less important.

0 views
Stratechery 3 days ago

2026.46: The End of Jensen

Welcome back to This Week in Stratechery! As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone . Additionally, you have complete control over what we send to you. If you don’t want to receive This Week in Stratechery emails (there is no podcast), please uncheck the box in your delivery settings . On that note, here were a few of our favorites this week. Presented by Agentic commerce purchases with Link, Stripe’s digital wallet that works surprisingly well for agents, grew 38x in September. The catalyst was consumer agents like Muse and Instinct. Filling out a web form is not a value-add activity for anyone. No one chooses to fill out web forms in their spare time. That’s why these agents have been particularly useful.  I’ve written about how these platforms are giving ordinary people an AI agent with a computer, enabling agents to do genuinely useful things on our behalf.  When one of those things is spending money, it needs a wallet that knows what it is, and isn’t, allowed to do. Stripe has learned that’s really about deciding how much freedom to give it. You might approve one purchase, or give it $500 for flights this week, and let it figure out the rest. The wallet’s job is to keep it within those lines while handling everything between “go buy this” and the purchase actually happening. That’s what Stripe is building with Link: You approve what it spends, it can come back to you if the final price changes, and eligible purchases can even come with protection if the agent makes a mistake. You can use Link on Muse, Instinct, as well as many other of the agents you might use already. This week’s Stratechery video is on AI on the Ballot . The Shocking Demise of Jensen Huang. This week’s coverage was dominated, unsurprisingly, by the shocking and sudden end of Jensen Huang’s 32 year tenure as CEO of Nvidia. The end came on Thursday morning, and just a few hours later, we recorded a Sharp Tech episode recounting Huang’s legendary history, figuring out what happened, and looked forward to what this might mean for Nvidia, the tech industry, and America going forward. Also see Andrew’s view from DC, Jensen Huang Was Always Going to Lose , and listen to Wednesday’s episode of Sharp China about the geopolitical angle, and Thursday’s Dithering about both the parallels and differences between Steve Jobs and Apple. — Ben Thompson Enterprise, Consumer and Intel.  The more things change, the more they stay the same: Intel remains on the back foot in the AI era, and the foundry dreams continue to flounder. On the heels of disappointing earnings and 2027 guidance that promises yet more letdowns, Ben’s Update on Wednesday examined the state of America’s would-be national champion by drawing comparisons to AMD, an Intel peer that’s red-hot right now. To understand why one company is capitalizing on skyrocketing enterprise demand while the other confronts flatlining consumer sales, the answer starts with chips chiplets and leads to more profound questions about culture, Intel’s nature, and its bifurcated strategy for the long-term (and, for subscribers, look forward to more on this topic in Sunday’s Sharp Tech mailbag episode).  — Andrew Sharp And Then There’s Professional Basketball. As a palate cleanser after a week of Jensen drama, this week’s Greatest of All Talk was a look at the top and bottom of the NBA after the first three weeks. On the euphoric side of the ledger we have Warriors fans (pre-Steph injury), Lakers fans, Spurs fans, and on the other end spectrum there is nothing positive to say in Brooklyn, a familiar malaise in Cleveland, and precisely one silver lining in Salt Lake City. All of it was a great time; we also hit Adam’s experience with the NBA in Mexico City last weekend, the hardest working players in NBA history, and an anxiety-inducing conversation about the number one pick, AJ Dybantsa.  — AS Jensen Huang’s Affair, The Culper Report, Huang’s Righteousness — Nvidia CEO Jensen Huang admitted to an affair with a Chinese spy, and that might not be the end of his troubles. I hope he survives, but his influence will never be the same. The Jensen Huang Story Gets Worse, Washington vs. New York, Silicon Valley — The Jensen Huang story got worse for him on Monday; the real risk for Huang is governance, even more than national security. Who benefits? Intel Earnings, AMD Earnings, Intel and AMD’s Divergence — Intel and AMD’s diverging earnings point to fundamental historical differences, and make me wonder if Intel needs to find a new (old) path. An Interview With Roblox CEO David Baszucki About Aggregating Play — An interview with Roblox CEO David Baszucki about how AI is transforming game creation, and why that opens up huge opportunities to both enable developers and make Roblox both an Aggregator and a platform. Jensen Huang Was Always Going to Lose — Jensen Huang spent four years substituting his own judgment for that of the U.S. government. His downfall was inevitable, and probably isn’t over. The Jensen Huang Affair Jensen Fired How Cisco Got Bullwhipped Indonesia’s Mega Rice Failure Jensen Huang Likely Out at Nvidia; The Megaspeed Maelstrom in D.C.; APEC in Seven Days; Legislation That May Come Next Agony and Ecstasy After Three Weeks, The “Earn Their Ice” All-Stars, Difficult Conversations About AJ and Season Ticket Holder Decorum Subscriber Mail: Twitter, Microsoft, Bending Spoons, Movie Reviews Nvidia After Jensen Huang, What Huang Meant to the AI Ecosystem, Washington Gets Its Revenge

0 views
Stratechery 4 days ago

Sharp Tech on Video: Nvidia After Jensen Huang, Ominous News for American AI, D.C. Gets Its Revenge

Sharp Tech is, for the first time, available on video. You have three ways to consume it: YouTube, Spotify, and here on Stratechery

0 views
Stratechery 4 days ago

An Interview With Roblox CEO David Baszucki About Aggregating Play

Listen to this post: Good morning, This week’s Stratechery Interview is with Roblox CEO David Baszucki . I’ve interviewed Baszucki twice, in February 2022 , and November 2023 . What struck me about this interview is just how much AI has changed the landscape for Roblox. This was a company predicated on the concept that anyone can make a game, but now, for the first time, literally anyone can make a game. That means a world of abundance, and in that world, there is an opportunity to be an Aggregator, and Roblox is aggressively pursuing that opportunity. To that end, we discuss how AI has changed Baszucki’s view of what is possible, what goes into a game creation harness, making games on your phone, and the power of recommendation. Then we talk about Roblox’s vision of a universal player, where a Roblox game works everywhere, and Roblox Everywhere, where Roblox developers can use the platform beyond the Roblox app. We also touch on advertising — the natural Aggregator business model — and whether Baszucki still thinks he is building the metaverse. As a reminder, all Stratechery content, including interviews, is available as a podcast; click the link at the top of this email to add Stratechery to your podcast player. On to the Interview: This interview is lightly edited for clarity. Topics: The AI Impact | A Game Creation Harness | Roblox Build | Roblox Discovery | Universal Player | Roblox Everywhere | Advertising | Metaverses David Baszucki, welcome back to Stratechery. DB: Thank you, Ben, for having me back. It’s great to be here. So this is the third time I’ve interviewed you, and I want to sort of put that in an interesting context. First, you started Roblox in 2004, you went public in 2021, you first talked to me in 2022 and again in 2023 , that’s like a nearly 20-year stretch. Now it’s 2026, and here’s my question: Is the world more different for you now, compared to three years before, than it was for the entire first two decades from starting the company to the last time we talked? DB: The changes in the world we all feel are sure happening a lot more rapidly. When we first published Roblox — I don’t know if people on this podcast remember — software used to be distributed on floppy disks and CDs. You know, the funny thing is, you’re the second person I’ve interviewed in the last two or three weeks where we’ve referenced back to fact that the reason we say packaged software is it used to actually be packaged! DB: It’s been crazy, we had the Internet, when we started the company, we had the vision of a category called immersive co-experience, sci-fi writers have been writing about the holodeck forever, and we went along our way. I’m still pro-metaverse, we can use the word metaverse, it’s okay. DB: Okay, good, we use them all. It’s Snow Crash — metaverse, holodeck, they’re all good. We started on this vision of building human co-experience, we saw the mobile age come across us, we saw the crypto age. The AI acceleration wave we’re in is faster and bigger and crazier than anything I’ve ever seen, it’s changing how we think about the product, it’s changing how people are going to create games. What’s really exciting is, when we started the company, we had this slogan called “You Make the Game”. DB: We started with that slogan, we learned along the way that the ratio of people who might want to play games to make games has drifted — more people want to just play games than make games, because historically making games has been Roblox Studio and 3D editing and writing script. We could be approaching the time where “You Make the Game” really comes true, and where making a game is accessible to everyone. It’s really fun, I’ll tell you a bit about some of the experiments we’ve been having with Roblox Build . But we’re entering an era where making a game could be as big as playing a game. I’m right there with you. We’re obviously going to touch on all those points, I think it’s probably predictable what we want to talk about. But to go back to 2023, you called AI then, “a super paintbrush and a super 3D modeler, rather than a full replacement” — Is that still how you’re thinking about AI? DB: It’s what the data is proving out. We just had our earnings call a few weeks ago , we highlighted a lot of acceleration. One, there’s various ways humans can augment with AI on Roblox, they can use MCP , which is like plugging in a model, they can use the Roblox AI assistant to help build. And what do you mean with that MCP server? You mentioned you had 5x growth quarter-over-quarter. That’s where you’re actually using Claude Code or Codex or whatever to work with Roblox? DB: People are plugging in, we can see all of them — Claude Code, Codex, other models — all being plugged into Roblox Studio to help build things. We’ve seen two games hit the top 20 made by AI, we have other metrics like the number of assets. What do you mean by “made by AI”? Because you also said 80% of games in the top 100 have AI-contributed content. DB: Touched by AI, vibe-coded directly with AI, and we can see the count of assets accelerating. DB: It’s turned out to be, I believe, somewhat faithful to what I said. This is a human paintbrush accelerator, good ideas are still very important, novel types of gameplay are very important. But people are able to flex their creativity much more quickly. We have members of the Roblox board of directors carrying a mobile phone around, building games like 15 times a day. We have people who are not necessarily engineers or product managers now building within the company, so we’re really at the dawn of a new era. You mentioned the super paintbrush, super 3D modeler, is there a super gameplay developer? Because there’s a large step between, “I have an idea for a game”, versus the actual mechanics of the game. How much can the AI do to help you as far as that goes? DB: You’re touching on the dream of what is a game creation harness versus a coding harness . Dive into it, this is super interesting. Lay out the difference and the way you’re thinking about that. DB: So everyone’s getting familiar with coding harnesses — Claude Code, Cursor, Replit, Lovable, Codex, you name it. A game is arguably much more complicated than code. As you mentioned, there’s art, there’s 3D objects, there’s scenes, there’s avatars, there can be gameplay code, there can be missions, and there can be code underneath it. So a harness that is being created to create games really has to run in multiple modalities, needs to listen to you, and if you said, “I would like a multiplayer tortoise-and-the-hare game that 100 people can play, and we’d like to do it on Venus, and I’d like it to run on all devices” — there’s a lot of different things that need to be generated there. The avatars, the assets, all of that, as well as the gameplay, which I think is a very human thing — your idea of what type of a game it is is actually very unique. So one of our board members is building a hybrid kind of Wordle-crossword game with his own vision of how this crossword game should work, and that’s very unique to him. I’m optimistic we’re still seeing that notion of a human coupled with an accelerating paintbrush in essentially a very complex harness for building games. There’s so many pieces you talked about. The asset creation piece, compared to 2023, much further along, arguably there. You could have even the video component, or the moving worlds, but then you almost have these different pieces. Do you have to build — is there going to be one harness, and one sort of — we’re going to get to your Build app in a little bit — that handles all of that? Or are you still thinking we need to do this piece-by-piece as the model gets good enough for, or we need different models maybe for the different parts of this? DB: I think just as we see with harnesses that are routing different models, even for coding now, a lot of companies have multiple models running side-by-side and their harness is optimizing. Our harness inside Build and inside Studio Build is running not just multiple models for certain things, but then there’s a second dimension of, “Are we building a scene?”, “Are we building a 3D object?”, “Are we building code?”, “Are we building an NPC?”. So yes, we are routing both by kind of specific asset that’s being created, as well as multiple models within each of those, and we’re combining some areas where we have incredible data on Roblox — for example, data that can be used to help create non-player characters, characters that act normally, side-by-side with you. But there’s also really good coding models out there, both open source and frontier, and so we see a harness being able to put all these things together. What’s the mix of those models? Say we have three categories — frontier, off-the-shelf open-source models, and models that you’ve trained specifically. And I guess even there there’s two categories, are you doing foundational models on your own, or just doing some post-training? DB: We literally have all three of those running in our Build harness. And we’re, in certain cases, optimizing the cost of frontier with the capabilities of open source. We have a Cube model — we’ve announced it — which is trained on 3D assets. Yeah, super interesting. Tell me about the Cube model. DB: The Cube model is our own 3D asset creation model trained on all of the 3D objects inside Roblox, where the developer has opted in to the training. It’s very fast, it’s very efficient, and it runs on our own hardware. When you go into Build or I go into Build and I say, “Hey, make that tortoise-and-a-hare game”, and if you describe that tortoise as somewhat unique and different, “I want the tortoise to look like this, and have a fur that looks like this, and maybe have this”, that Cube model will and may come into play and generate that tortoise for you. How does your cost structure work for this? Is it a usage thing from a developer perspective? You have the MCP server — that’s where they can bring their own, what’s the mix? DB: You sound like a product manager, because these are exactly the same discussions I had about an hour ago. The ultimate vision is something like Roblox Build is just like the creation of short-form video. Short-form video 10 years ago — all of a sudden we all figured out we can make video and edit it on our phones, amongst multiple products that was like a huge revelation. We are so on the same page, I have a whole section on short-form video later in the interview. That’s good foreshadowing. DB: Yeah. And so along comes short-form video, and we have phones, and all of a sudden we’re making and sharing video. Arguably games are a lot more complicated, they can be immersive 3D environments, multiple people can come there. They have, as I was saying, code and geometry and avatars, but that’s where Build is going — that same notion of, “I’m going to pull out my phone and create something”, some of those models are still expensive. But the dream is, we’re already testing — I think Canada went live today. We’re testing in, I think, Serbia and New Zealand, we want it available for everyone and so it creates a wonderful opportunity for us to be super efficient with our models while super powerful. It creates the opportunity for those users who want to just do hundreds of iterations to have them buy tokens as well, if they really want to dig into it, so we’re optimistic this will be free for everyone. You keep mentioning Build, I’m on your website , and this is how you describe Build: “Build is a new way to make your own Roblox game right from your phone. Just describe the game you want in your own words, and Build’s AI creates a playable game for you in real time. You don’t need any coding or game-design experience – if you can describe it, you can build it. Want to change something? Just tell Build what to adjust, and keep going until it feels right”. Is this what you always wanted? Are we going to look back and say Roblox Studio, your pride and joy, was just a stepping stone to this? DB: This was a sci-fi fantasy when we started the company, and arguably a fantasy we never thought would come true. I would say when we started the company, we couldn’t have imagined it, and for many years, I think many people in the industry were always trying to figure out ways to make editing 3D easier, “Oh, that’s a really easy consumer 3D modeler” — even the best consumer 3D modeler is still pretty complicated. It’s unbelievable, my current 3D experience right now is, I’ve gotten very into 3D printing and I’ve never opened the associated app for it at all. It’s all done in the AI and just goes straight to the printer, it’s unbelievable. I appreciate all those people who did all the actual modeling and stuff for the last few years — I’ve skipped over all of it, but it’s the same thing in digital space. DB: I would say in the last five years we started to see that maybe AI can do this and I would say it’s been in the last two to three years where the rate of acceleration has started to surprise people and as that rate of acceleration has started to surprise people, this dream becomes much more possible. There are games on Roblox right now that are earning money that have been made with Roblox Build. We have not tens or hundreds, but thousands of games that have been published in Roblox Build. It’s not live in the U.S., we’re testing, and it will be soon, so the data is already out there. What’s so interesting is you saying, “We couldn’t have imagined it”, because one of the things in our previous interviews that came across again and again was the continuity of your vision and like, “Yeah, we saw this back then, and now we’re getting there, and we’ve had this vision all the way” — and that sort of goes back to my first question. Is there almost been a break in the last few years where even you are like, “Holy cow, I didn’t know this is where we were going”? DB: I feel like all of us in the technology industry are feeling that — everyone, CEOs, programmers, people who use these products, consumers. We’re all in the middle of it, I would say, over the last several years. Do you think this makes Roblox more or less valuable? If anyone can make a game, is Roblox Studio, your jewel, not that big of a deal anymore? DB: I think what really ends up being valuable is the combination of a few things. We’ve shared the notion of Universal Player, that I’ll dive into, the notion of Roblox Everywhere , the notion of the power of distribution platforms becoming more and more powerful when a game is created, “Where do I find it?”, “Where does it run?”, “How do we keep it safe?”, “How do I play it with my friends?”, “Is there an economy?”. So in a way, distribution platforms become more powerful. It’s worth thinking about it like — pick a distribution platform, whether it’s TikTok or YouTube, in a way, AI can accelerate those platforms, but in a way, those platforms have enormous strength with or without AI and even though someone could reproduce the whole YouTube platform with AI — the cloud, the client, all of those things — people are still going to go to YouTube to watch video. Yeah, this is the connection I was drawing — going to the short-form video platform idea. I’ve always drawn a distinction between distribution and discovery , distribution is free — anyone can post a website, at least generally speaking, they could host a game on their website, they could host a game on Vercel or wherever it might be. But then discovery is, if there’s a gazillion games, how do you find what you’re looking for? Of those two, what do you think is more important for Roblox’s future? DB: They’re both enormously important. The Roblox platform itself has well over 100 million people coming every day. On the discovery side of our platform there’s two angles to it, the first is, what happens when everyone’s building a game? All of a sudden, in the old days, there’s an abundance of new games and experiences. How do we connect the right person with the right game with their friends? Even in what might be called AI slop, there can be amazing games embedded in that AI slop, even if it’s 1 in 100 and even when games are crafted by hand, there can be some of those games that just aren’t interesting to people, so discovery becomes enormously important. What we have going on — and with the opportunity going on at Roblox — is that in addition to traditional discovery, which is optimizing every user-game pair and predicting long-term retention from that connection, with more and more games, content understanding becomes incredibly important and because Roblox is a platform where all of the games are running in the cloud, we can examine the content, learn from what we’ve learned, and maybe say, “This game has an interesting, new, diverse way it’s been constructed, this game looks a lot different than other games”, and that can help surface great content to new users. The other thing, in addition to on-platform discovery, is this notion of Roblox Everywhere that we’ve shared. Let’s get to that, I promise we’ll get to it, because I think it’s super interesting in this context. But you’ve gotten much more into — you mentioned matching players to the right game, what have you learned from getting into recommendation algorithms? The path — I keep referring to it — but this is the social-networkization— DB: Yeah, it’s very different than video, because a game is something a user can play multiple times, a game is an environment where friends can have fun. Some good games last for years and years. You know, Adopt Me has been on the platform forever, so finding an Adopt Me that creates joy for a long time is very, very important, rather than a one-shot video. I think what we have learned is to optimize for longer-term retention rather than short-term retention. Well, that really hit your earnings when you made that shift a couple quarters ago. What was your thinking? Is that what you expected? DB: I don’t want to comment on hitting the earnings, but I want to comment that mathematically, if we can raise long-term retention — which means more people stay on the platform, more people have a good time, and, like I mentioned with Adopt Me, find games they might play for 10 years — that pencils out. And retention is critically important in growing users and growing word of mouth and growing all of those things, so we’ve gotten better and better at optimizing long-term retention. We measure it directly, we measure the contribution to long-term retention from every user-game pair. We see interesting things — different types of groups of people like different types of games and I think if we go to the Roblox home page, for example, and I play, or someone else plays, I think people are universally saying, “Oh my gosh, those are more and more looking like games that I’d like to play for a long time”. You told me in 2022 that Roblox’s initial product failed because it wasn’t multiplayer and social and that was one of the foundational lessons of the company. Now you’re enabling offline single-player games. Have you moved? Is this what has changed? What’s shifted, where — yes, of course you’ve mentioned the social aspect multiple times — but also there are games people just want to play, a great game, on their own? DB: I feel long-term the foundation of Roblox will continue to be multiplayer, friends coming together, that may be in real time, that may be asynchronous, that may be “I take a turn, you take a turn”, that may be I play a match-three puzzle at some point and compare it. What the move to single-player, the move to puzzle, is really saying with our aspiration for Universal Player and Roblox Everywhere, we have enough bandwidth now to cover the whole spectrum. When we talk about the notion of Universal Player, the notion is going all the way from a traditional 2D offline puzzle game, all the way to that which is yet to be invented — which is super high-scale multiplayer, photorealistic reality — and everything in between. So the vision with the puzzle game, as well as Roblox Reality, is going back to the video analogy. In a way, with video, the camera and the player have been somewhat invented, we have cameras all over, we have 4K, 60-hertz playback, we have playback mobile, we have playback in our living room, and so all of the video out there generally will play back on my living room 4K, 60-hertz TV. That’s not true in gaming. There’s not a universal rail, where every game as of yet invented, as well as to be invented, you could imagine it playing there, so that’s the notion of 2D puzzle and Universal Player. Presented by Agentic commerce purchases with Link, Stripe’s digital wallet that works surprisingly well for agents, grew 38x in September. The catalyst was consumer agents like Muse and Instinct. Filling out a web form is not a value-add activity for anyone. No one chooses to fill out web forms in their spare time. That’s why these agents have been particularly useful. I’ve written about how these platforms are giving ordinary people an AI agent with a computer, enabling agents to do genuinely useful things on our behalf.  When one of those things is spending money, it needs a wallet that knows what it is, and isn’t, allowed to do. Stripe has learned that’s really about deciding how much freedom to give it. You might approve one purchase, or give it $500 for flights this week, and let it figure out the rest. The wallet’s job is to keep it within those lines while handling everything between “go buy this” and the purchase actually happening. That’s what Stripe is building with Link: You approve what it spends, it can come back to you if the final price changes, and eligible purchases can even come with protection if the agent makes a mistake. You can use Link on Muse, Instinct, as well as many other of the agents you might use already. So step back and explain Universal Player, broadly speaking — both in terms of the different surfaces, and also bringing in games from other platforms into the Roblox experience. DB: Part of this comes — I’ll come at it first from an AI angle, and then a puzzle game angle. From the AI angle when we hook either our Studio assistant or an MCP model up to Roblox Studio, and you or I say, “I’d like to build this”, there’s still technical constraints, if I hooked up MCP and said, “Build a game that happens in the state of California, make it a 3D multiplayer environment, I’d like it to be photorealistic, and I’d like it to be able to have up to 100,000 people all walking around and driving cars” — not technically possible. But we do want to push towards that — once again, multiplayer high-scale photorealism — as quickly as possible, and make a place for anyone who’s either imagining a game or building it with MCP server, or, as you mentioned, ultimately bringing in a game they maybe have already made and converting it to that type of environment. So if we have a Universal Player, that’s a superset of all games that can run multiplayer. You could imagine that if you had a puzzle game that’s offline, connect it up, bring some of the source code in if you want to, and we’ll convert it to Roblox. Or you had a single-player high-res game, we’ll convert it to Roblox and in that conversion process, a couple other cool things will happen. We coined the term “Robloxification” — it’s really what’s different about the Roblox architecture. Your single-player game from a PC can now be multiplayer by default, because we’re running against a cloud. Because we’re doing 3D streaming, your single-player game will also run on a phone — oh, it’s being Robloxified. Your single-player game, rather than being a 500-gigabyte download, you can just join it immediately, that’s very interesting. Your single-player game has access to economy and other things so the notion with Universal Player is really both convert, as well as Robloxify, existing games as well as new games. Is this leveraging — to go back to the discovery versus distribution — is this leveraging the fact that you want to get into Roblox’s discovery engine? There’s people on Roblox, they’re not going to go download your game individually, they’re not going to go to another website. DB: That’s exactly right, in that a game brought into Roblox Everywhere in our Universal Player will then be available in the Roblox app. And as we thought about it, what a great way to discover more content than to have great games out there that are also discoverable in the app. This is the analogy — to go back to the short-form video — where you had this story with Meta versus TikTok. TikTok realized that it’s called a social network but it’s not really a social network, it’s just really giving you the videos that it thinks you are going to like, and who you follow doesn’t have anything to do with it. Meta is stuck in their social network, “That’s what makes us powerful”, and to compete with TikTok, to make Reels into a thing , they had to let that go, “We’re just going to give you what you want”. Is there an analogy there to your embracing single-player games, to you having recommendation algorithms? And sure, Meta will tell you the social network part is important, will always be important. Is that analogous to you saying, yes, the social aspect of Roblox is important, will be important, but actually to truly grow, to reach your full potential, you have to actually atomize a bit more and meet the player with exactly what they want? DB: Yeah, I think if anything it’s even bigger than that, I’ll go through three realizations. The first realization that we’ve come to, looking at the physical world, when people play — whether they’re playing sports or a board game or any form of play, jostling around with friends — it’s typically not just play, it’s playing, it’s connecting, it’s community, it’s creating the play, “What are we going to do?”, it’s sometimes watching other people’s play, it’s dressing up to play. And so we started thinking about the market, not as the game market, but the play market, and how all of those pieces really work together so that’s been a bit of a stretching, rethinking things for us. The second is, when we say the play market, we can no longer just say, “Well, Roblox is a 3D multiplayer game engine”. You were constraining yourself. DB: That’s right. All of the constraints, I think, in the past that helped us focus and grow at a certain scale have to flip from focusing constraints to constraints we give up on. What was more difficult — convincing the company we have to let those constraints go, or did you have to convince yourself? Because again, two things that really stood out. Number one, you had this long-term vision, and it’s actually been continuous, and now you’re saying, “Wow, this has been a big shift”. But then number two, this being focused, “This is what we do”, and now you’re like, “Well, actually, no, we need to let that go and do everything”. DB: We’ve gone through several of these. The early version of Roblox was PC and Macintosh. We, for a while — if you can believe it — there was no mobile phone version of Roblox and when we looked at the iOS App Store top 100, there was no 3D game in the top 100. China was a little bit ahead, they were starting to poke at it, the U.S. was not. Giant company debate, “How could Roblox ever work on a phone? The phone is 2D. It’s not going to work”. Luckily, mea culpa, we went for it and now we’ve gone through that transition. There’s also a time on Roblox, very early on, we thought the blocky look was so simple and easy to build with, we wouldn’t allow the importation of 3D mesh objects, we mea culpa’d it, because ultimately we want all styles of game. And so I think 2D and offline is kind of one of the last holdouts for that, in saying it’s really the play market, and we can’t constrain that anymore. Are you pushed into this market, or are you pulled — just the market opportunity is so large? DB: I think it’s so much more fun and simple to just say, “We’re going after the play market”. (laughing) That’s right, let’s do it all. DB: We need integrated and safe ways for community, we need integrated and safe ways to find great games to play with, we need all of that just in a simple — it’s actually a simpler product spec and a more fun product spec. Does that mean, you as a company, you’re more defined by — maybe you’re like supply-defined, where, “What can we do, and this is what we’re going to deliver”, and now that’s sort of flipped to, “What’s the market and how can we fill that market?”. DB: Yeah, I think it leads to all kinds of interesting features of the product. For example, short-form video is traditionally used for watching by yourself. Imagine something that feels like short-form video, where you’re seeing the 3D game where your friend is already playing, and it’s a way to find a friend and jump into a game. Right. We made the short-form video comparison, but you’re literally doing short-form video. DB: Finding 3D ways to discover is super interesting, finding ways to connect with your friends is super interesting. I feel this singular vision on play is a category that’s got legs and I think the thing that is yet to be figured out is how big is creation versus playing with friends. What I have seen with the early players on Build is, there’s a lot of joy in making a short-form video on one of these platforms and having your friends see it. That happens in spikes and I’ve seen people do that. The joy I’ve seen of people creating games on Roblox and literally sitting there with their phone and saying, “Oh my gosh, I have four people in my game, and those four people are playing my game, giving me feedback” — I’ve seen that be really powerful. That’s interesting, because the feedback mechanism — you can act on it. If someone gives feedback on a video, it’s just a comment on a video. DB: And like I highlighted, board members doing 15 modifications a day to their game so that four people turns into six people. That’s really interesting. It’s like there’s a 3D aspect to gaming, as opposed to the 2D aspect of a video. This is almost like a 4D aspect, where the interaction — it’s a social network in a very different sort of way , it’s around this object that could evolve and move with the interaction over time, I think you do have a 4D product. DB: That’s right. Well, we use the term 4D lovingly in our physics engine, because 4D objects are objects that have behavior and code and do things, rather than static objects. The other topic you’ve brought up continuously is Roblox Everywhere. First, explain what it was — I have some questions about this that are connected to what we’ve been talking about. What is Roblox Everywhere, from the top? DB: We shared earlier, our jump from PC, Mac, “Oh my gosh, we’re doing mobile, could it work?” — and then we shared some of the other changes we’ve made along the way. One of the things, for quite a while, we’ve held on is: 100% everything is in the Roblox app. DB: But as we’ve seen our vision of the Universal Player — the vision that, whether you build on Roblox or you convert or you use AI, below the rails, Universal Player is the place a game can be constructed and run at scale — as we thought more and more, all of a sudden the power of what is the Universal Player becomes appealing, not just in the Roblox app, but possibly for a creator who wants to ship a standalone game in the iOS App Store, the Google Play Store, on their PC, on a console. And as we thought through it, the benefits of giving creators access to the Universal Player, even when they’re not in the app, brings an enormous advantage to us, which is that their content can then be available for discovery in the app. That gets to my question. Okay, I’m on board with the releasing, let’s be on all devices, let’s not just have blocky characters, let’s have 3D characters, all the things we’re talking about. But I’ve been framing this whole discussion — we’ve been comparing to short-form video, which is an Aggregator play. You own the end users, you own the consumers, you have Universal Player, people can bring their games in to reach the customers, you have an advertising business that we talked a lot about last time, I have some questions about. Roblox Everywhere seems to be like the whole company is now flowing in one direction, you have the Roblox place — if you want to make a game, go there, there’s customers there, there’s monetization there, it’s all there. And now you’re like, actually, let’s open up a trap door and let everyone go out the side, am I the one that’s now being the wimp and saying, “Can you be a little bit more focused”? DB: I appreciate your insight there, but what’s really exciting to us is the potential to get such a large percent of the gaming market running on Roblox Everywhere, and the potential content explosion that happens within our app. The way we thought about Roblox Everywhere is, if you’re a creator — you might be an indie creator just pushing a small app, you might be a large studio — the more we can have both the most amazing technology platform, really Universal Player, and the more it makes absolute sense for you to run on that technology platform, whether you’re in the Roblox app or standalone, we get the benefit of that content in our app which is social, which can be launched instantly, which shares a common social structure, which carries a common safety structure, a common economic structure, all running together. So our calculus — and as you said, maybe it looks risky or whatever — but our calculus is actually, this is the best way to grow users in the Roblox app itself. Yeah. Well, it’s interesting because I think this gets on the — I’ve been drawing the analogy, we’ve been drawing the analogy jointly, between short-form video and gaming. But to your point, gaming is still a lot more difficult and if you go through the history of gaming over the last 40, 50 years, this has been a trend is over time, just the cost of developing gets so large, you need to be multi-platform. The PS3, one of the all-time examples, builds this super complicated, powerful architecture that no game ever leverages, because the cost of creating HD assets is so much higher . They need to be on Xbox also, they need to be on PC, and so everything goes multi-platform, Sony responds by doing the exclusive strategy, etc. Is this a similar articulation, that making games is hard, and so if we don’t make it so developers can run their games everywhere, they’re not going to be on Roblox either? DB: It’s really hard and right now there’s a bifurcation, a lot of times, between that super cool high-end PC game I see and what I can run on my two-gigabyte Android phone overseas. Those typically are different products. And so you’re trying to make the whole world the market for a developer — if you build on this, the whole world will be your market? DB: Imagine if it was technically feasible to build for super high-end AAA multiplayer on a PC, and connect to the same game on a two-gigabyte Android phone overseas. That’s a technical opportunity, we think, for creators. How is that different from streaming? Like, you stream a console game to your phone, what is different about what you’re proposing? DB: There’s a very creative history of what is called game streaming, where the game is run far away, I send my keyboard and mouse to the game, and then video comes back. (laughing) Right. And it sucks. DB: And this was even 20 years ago — Gaikai, OnLive, OTOY — we’ve seen a long history of this. You said it sucks, the way I would say it is, some of the things we’re doing in our video — you and I are on a Zoom call right now, when I look at my character, my character’s reacting instantly, I feel like I’m looking in a mirror, in a pure video streaming architecture, I might move 50 or 100 milliseconds later than when I do that, and people are very sensitive to that. If you’re close to the video server, maybe it’s pretty fast but I do believe having local 3D instantaneous feedback is super, super critical. So that could be one reason why video streaming just hasn’t kind of exploded as much as we thought. Is there an issue where streaming — they’re trying to retrofit what existed, and you’re natively building for this experience, and that makes a difference? DB: We want native 3D, you know, we want multi-device native 3D, very low latency happening on those. There’s some other dimensions where gaming is getting more difficult as well. The regulatory environment — it’s not like the old days if you or I make a game. There are different laws in every state and in every country, cultural laws, all of those. One of the opportunities is to have a platform that can automatically adjust to that — what are all of the rules around age, around ratings, around all of those things? That’s another thing that can help a creator reach the world without having to think about all those things. How do you think about monetization for the Everywhere games? Is there going to be like a platform fee, like a Unity sort of aspect? Or is the payoff, “If they’re building using our technology, they will also be in Roblox, and that’s where we’ll make the money”? DB: We shared at RDC the notion that monetization may be somewhat similar for games Roblox would publish. And so if you went to an app store, there may be some games published by Roblox that are the exact same games that are inside Roblox. We would only do that with the developer’s permission, but it could be really cool — you could go to your app store and search for “Grow a Garden” and say, “Oh my gosh, there’s the game”. For games where the developer is pushing directly to the app store, we haven’t shared the economics but in the thesis of what I said — which is eliminating any reason for the creator not to think about it — that leads to the economics needing to be pretty provocative and so stay tuned on that. Where is advertising fitting in this and your vision overall? Give me the advertising update. We spent a lot of time on that three years ago. DB: We shared on the earnings call that we’re more and more excited, and we’re all in, not just on Roblox platform advertising, which is growing, I think we shared that’s growing very quickly as well. As creators come into Roblox and use Roblox Studio Assistant or Build or MCP, “Okay, I need to get some initial traction, I need a thousand users”, those ads are available on the Roblox platform. And isn’t that the great thing if it’s abundance, there’s too many games, you have to discover it — user acquisition ads are very profitable. DB: We have seen an acceleration of that type of advertising and to take that all the way back to the prior analogy, we have board members on their phone buying advertising. (laughing) Conflict of interest, they’re bumping up their own stock price. DB: There’s a second type of advertising that, more and more, is becoming available, which is very short, five-plus-second skippable pre-roll video. There was a time, way back in the history of Roblox, we tried pre-roll video, it was not a good experiment. The quality of the video was not super exciting, pre-roll video at that time was 30 or 45 seconds. Technically it was a pain in the butt, and we yanked it out because we didn’t like the user experience. That kind of left us with, for a while, “Uh-oh, no pre-roll”. But today, the fun and the quality — if we’re very selective of the brands, and we could imagine the brands that are interesting to a 21-year-old Roblox player — clothing brands, electronics brands, movie brands — where they might actually say, “It’s pretty cool to see this”. We also, because typical games on Roblox launch not quite instantaneously — they launch maybe in five or six or seven seconds. Right, you have time. DB: There’s actually a beautiful technical overlay there, if we do it right and do it natively, to put this all together so we’re very excited about this potential combination. Is there an aspect where companies, especially when they’re first doing advertising, they overthink it? Like, “We have to have the world’s most inventive ads, there’s going to be billboards in the game, and these experiences” — and actually, no, just show a pre-roll video, it’s fine. DB: I think that’s right. I think we have a history of thinking big, and I do feel ultimately there’s going to be a wide range of types of branding that people do. There’s a lot of immersive 3D potential for literally shopping in 3D and doing all of that but there’s also an enormous quantity of pre-roll video. Right, you can get to that eventually, there’s actually a lot of low-hanging fruit, it feels like a lot of first-time advertiser platforms — they’re reluctant to pick, and then they’re like, “Actually, let’s just pick the low-hanging fruit and go from there”. DB: I think that’s right and I think in a COPPA -compliant, privacy-compliant, Roblox-values-compliant way, we can still imagine amazing user experience for advertising. Does this tie into the potentially provocative pricing for Everywhere? Where you make a game, it’s everywhere, congratulations — but if you want to actually make money, you’re going to need to be on Roblox and that’s the best place to spend money to acquire users. It’s better than spending on Meta, better than spending on wherever else it might be. DB: You could imagine, in under 18 on Roblox, we’ve shared we may have literally 20% of gaming revenue in the U.S. flowing through the platform and so that’s huge penetration in that age group. As we move towards that same penetration over 18, all of a sudden the value of advertising on Roblox for my new game becomes more and more valuable, because those users are already there. Are older players like your great white whale? I think every time that we’ve talked over the years, it’s always been, “We’re going to expand into older players”. When is that the reality, as opposed to the opportunity? DB: Reality is right now, the fastest-growing thing on our platform is 18 and up, it’s growing very fast. Are you getting new users, or is it people aging into and sticking with Roblox? DB: Absolutely, both. It’s a smaller cohort than the under-18 thing. But as it becomes an increasingly larger percentage of our revenue, it will more drive the growth rate of the top end as well. And how does that relate to the single-player offline, new types of gaming, puzzles, and all the things you’re talking about? DB: I think there are some of those segments that are very much 3D multiplayer, some of those segments are puzzle. There’s an overlap in that, as we think about the whole play market and we don’t restrict any types of games going forward, we have more opportunity to really create a good chunk of that market and once those users are on our platform, they’re very valuable, because other games can be attractive to them as well. In our first interview you talked about how you deliberately avoided paid acquisition, because organic virality was so essential to Roblox, in part because of the social aspect of the games. DB: I think we were careful with it. It’s a much bigger thing now, right? DB: For under 18, even in our early days, we experimented with paid acquisition, but we are very careful with always getting a return on the spend, and always knowing that organic viral had to be the biggest, but there still is a logical amount to spend. And we’re doing the same on over 18 right now, we are mixing organic with paid, there’s an optimal mix of them. We always want whatever we spend on acquisition to make incremental revenue more than what we spend and we are sticking to that hygiene. So it does make sense for us to do the paid acquisition. Is Roblox a metaverse? Is that still the dream? I like the word, no one else does. DB: I feel it’s weird how that word kind of came and went a bit but I think over time, maybe that word will come back, right? Because in the ’80s, ’90s, and 2000s, that was a very popular way of describing not just a gaming platform, but a communication platform — a place where people came together to play or to work or to do a bunch of things. I do feel ultimately, beyond gaming, that still makes sense. This Zoom call, at some point, will be a 3D simulated metaverse call, and we’ll be able to run in traditional 2D mode, and we’ll be photorealistic but if I wanted to go and look at that globe behind you, and I asked your permission, I said, “Ben, can I come and see the globe?” — you’d say fine, hopefully, and I’d come and look at it. And that’s something we can’t do on Zoom that will be unique to, ultimately, a metaverse call. It is a very cool globe. I got it at an antique shop. It has, like, a barometer in it, a thermometer, and it actually lights up — it is a lamp, also so it’s my background. DB: I love globes, I love telescopes too. Well, there you go. With the metaverse, with this 3D world, you just talked about a real-world thing, you could come into my room. But to go back to the example, you mentioned the developer that wants to build the state of California and driving cars around and 100,000 people, is this ultimately going to be realized, not through polygon drawing, but it’s all just going to be AI-generated ? DB: I don’t think it’s going to do that, and the reason we’re building this Roblox Reality architecture — and I’ll share a bit about it — is we do want, “What’s the quickest way to get to multiplayer, photorealistic, immersive gaming?”. Multiplayer is where the fun part comes in, and it’s why we talk hybrid. So in the movie business, there’s several famous directors right now who are going hybrid and what they’re doing is, they’re either shooting live actors, or they’re building a very rough scene for their movie in a 3D system, then they’re taking the 3D information and they’re upsampling it with an AI model to 4K photorealism and so they’re putting together the best of traditional — and in the case of live actors, there’s much more accurate capture of the face, of the emotion, when shooting real film for the actor, but then you can decorate the set however you want. In gaming, synchronizing 10,000 players needs to be done in the most efficient way possible and doing that in a video latent space is not the most efficient way, doing that in a rough 3D representation is the most efficient way. But for every one of those 10,000 players, it’s getting a pretty efficient 3D representation, you can imagine each one of them uniquely getting an AI video model upsampled for them specifically that looks like 4K, photorealistic 60 hertz, and I think that’s where it’s going to come out. Are you going to win on doing this because you have a technical advantage, you pull it off, or because you already have the audience? Lots of people could build this out. DB: I would say yes and yes, hopefully. This is a vertical integration play, the vertical integration. Right, which has always been a core part of Roblox’s theory. DB: I’ve played with, for example, Grow a Garden, or classic games on Roblox, with early versions of photorealistic upsampling, and it works out of the box. There is a notion that existing games can run for all players, same gameplay, but for some, photorealistically upsampled. But I also feel it’s a technology play, because most of the video models that are being built today are either single-player world models, or video generation models — like you or I want to use one of these models to make a cool 15-second video. This technology is different, this is real-time, low-latency video being generated from the 3D video and the 3D depth map and a bunch of other things, so it’s a very specific type of video model that has some overlap, I think there’s opportunity both for technology as well as vertical integration. The thing that I’m taking away is — and this has always been the case talking to you — the scale of the ambition, but it does feel like the big shift is actually, “Let’s literally do everything, let’s do the supply side, we own the demand side, let’s do all kinds of games”, is that what Roblox is today — we’re all gaming, period? DB: I would say, do everything within the carefully thought-out boundaries of the future of play. And so within play, yes, let’s be open to 2D puzzle, let’s go all the way from two-gig Android to photorealistic Roblox Reality and in between, let’s make things available in the Roblox app as well as in Roblox Everywhere. So I would say, within those hopefully thoughtfully constructed guardrails, then yes — in a way, it’s do everything. Do you feel okay without the focus? Without knowing, “This is what we do”, is, “We do everything” — is that focus in its own right? DB: It is focused, because there’s still some pretty tight guardrails. And so I think, given the size of the company and the ability to use AI to accelerate engineering, I think it’s the right current product spec — just like when we started, it was PC-only, blocky, and that was good enough. Last question, you mentioned the AI component. This “do everything” — is that downstream of, “Now we can use AI to program”? Or did the AI programming come on just in time to make this vision possible? DB: For many companies, I think like ours, back and forth, iterate, iterate — AI engineering acceleration, expanding maybe the definition of the product spec, AI engineering, expanding the definition of the product spec. But I do think companies around the world are now starting to take things that they thought would be a nine-month project, and thinking of it as a two-month project. How do we rethink how we architect, to make these nine-month projects two-month instead? Very good. David Baszucki, very interesting, very cool to see what you’re doing. DB: Great talking with you. Thank you again, Ben. This Daily Update Interview is also available as a podcast. To receive it in your podcast player, visit Stratechery . The Daily Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly. Thanks for being a supporter, and have a great day!

0 views
Stratechery 5 days ago

Intel Earnings, AMD Earnings, Intel and AMD’s Divergence

Intel and AMD's diverging earnings point to fundamental historical differences, and make me wonder if Intel needs to find a new (old) path.

0 views
Stratechery 6 days ago

The Jensen Huang Story Gets Worse, Washington vs. New York, Silicon Valley

The Jensen Huang story got worse for him on Monday; the real risk for Huang is governance, even more than national security. Who benefits?

0 views
Stratechery 1 weeks ago

Jensen Huang’s Affair, The Culper Report, Huang’s Righteousness

Nvidia CEO Jensen Huang admitted to an affair with a Chinese spy, and that might not be the end of his troubles. I hope he survives, but his influence will never be the same.

0 views
Stratechery 1 weeks ago

2026.45: Falling Leaves (and Stocks)

Welcome back to This Week in Stratechery! As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone . Additionally, you have complete control over what we send to you. If you don’t want to receive This Week in Stratechery emails (there is no podcast), please uncheck the box in your delivery settings . On that note, here were a few of our favorites this week. Presented by I spend a lot of time trying to figure out what is happening to businesses on the internet. What I often find I’m missing is a view across the economy itself. With more than 1.7% of global GDP running on Stripe, working with them means I can bring some of what they’re seeing to Stratechery. They are, in a sense, a barometer of the economy. Take the “SaaSpocalypse”. Earlier this year, nearly a trillion dollars in enterprise software value disappeared amid fears of AI eating its lunch. Stock prices only tell us what investors think will happen, and not necessarily what is happening. In this case, Stripe recently built a SaaS Index using payments data from 72,000 non-AI SaaS businesses. Looking at their actual performance, we can see that while SaaS stocks were selling off, revenue growth was actually accelerating. Year-over-year growth is now over 30%, with some of the fastest growth coming from the youngest SaaS companies. I suspect there are a lot more interesting things hiding in Stripe’s data, and I’m excited about having an excuse to go look for them. This week’s Stratechery video is on The End of the Internet . AI on the Ballot. This week’s elections delivered Democrats the largest Congressional majority either party has enjoyed in 36 years, and coupled with a variety of similar results at the state level (particularly in Texas), not to mention plenty of data center opposition among Republicans, the future of the AI buildout is now fraught with more regulatory uncertainty than ever. I wrote about X, 2024 symmetry, and the more political aspects of the election for Sharp Text , while Ben’s Article on Wednesday built on his data center warnings from April and provided 10 takes on the post-midterms landscape for tech companies. — Andrew Sharp A Google State of the Union. Last week Google reported earnings and saw growth in search revenue slow for the first time in five quarters. Given the appearance of mortality in what might be the greatest business in the history of digital commerce, investors were predictably spooked. However, I think the company deserves time to see if AI Mode ends up like Stories and Reels did for Facebook: if anyone can monetize AI with ads, it’s going to be Google. On Sharp Tech , we also got into why Google’s AI bet is about more than AI Mode, and why I find myself impressed by Google’s fortitude, even if the rest of the Internet is a casualty. — Ben Thompson And What About China? Elections have consequences, as they say, so what might a Democratic Congressional majority mean for the next two years of U.S.-China relations? We tried to answer that question on Sharp China this week , including predicting factional foreign policy fights between the left and center-left, highlighting the profound unknowns concerning President Trump’s posture after this week’s defeat, and offering one safe bet in the new environment: life is about to get a lot more complicated for Jensen Huang in Washington. — AS Google Earnings, Facebook Déjà Vu, The Agent Play — Google’s earnings miss looks a lot like Facebook’s, and that’s a credit to Google. The company is being more aggressive than they get credit for. Apple Earnings, Memory Problems, The Hardware Company Regression — Apple’s earnings suggest that the company’s hardware business is doing better than ever, even in the face of memory challenges. Does the software story help or hurt? AI on the Ballot — Data Centers were on the ballot, and they lost in a major way; what does it mean if American opposition to building is permanent? An Interview With Intuit CEO Sasan Goodarzi About Accounting and AI — An interview with Intuit CEO Sasan Goodarzi about the working for one of tech’s oldest software companies, and how Intuit is leaning into the AI paradigm shift. X Marks the Rot — Republicans colonized Twitter and then cannibalized their support with two years of memes, corruption and excess. DoorDash and Google The MacBook Pro (with touch) The Rise of GLM and China’s Zhipu AI Parsing the Fifth Plenum; US-China After the Midterms; New Scrutiny for Nvidia?; Cybersecurity and Espionage The Rapture and the Wild West, Luka Doncic on Praise Probation, AJ Dybantsa Saves the Wizards Subscriber Mail: Apple Past and Present, Omarchy, Instacart, and Baseball America Votes Against Data Centers, Implications for Big Tech and Beyond, Were Google Bears Right All Along?

0 views
Stratechery 1 weeks ago

Sharp Tech on Video: The U.S. Votes on Data Centers, Implications for Big Tech, Google’s Response to AI

Sharp Tech is, for the first time, available on video. You have two ways to consume it: YouTube, and here on Stratechery

0 views
Stratechery 1 weeks ago

An Interview With Intuit CEO Sasan Goodarzi About Accounting and AI

Listen to this post: Good morning, This week’s Stratechery Interview is with Intuit CEO Sasan Goodarzi . Goodarzi took a very non-traditional path to Silicon Valley, joining Intuit in 2004 after working at Siemens and Honeywell. After managing business units across the company he became CEO in 2019, and is taking on the challenge of figuring out how one of tech’s oldest software companies manages the AI paradigm shift. In this interview we discuss Goodarzi’s background and the what it takes to be an effective generalist, and get his take on how Intuit has thrived for 44 years. We discuss the shift from Quicken to Quickbooks, the TurboTax business, and how Goodarzi thinks about the consumer space today. Then we talk about accountants, Intuit’s most important customer acquisition channel, and how the company wants to change the profession from partners to customers via AI. I present the case to Goodarzi that Intuit seems like its being disrupted, and he makes the case why they are well positioned to actual grow their market. We also touch on acquisitions like Credit Karma and Mailchimp, and how Intuit has managed business model changes in the past, and what needs to change in the future. As a reminder, all Stratechery content, including interviews, is available as a podcast; click the link at the top of this email to add Stratechery to your podcast player. On to the Interview: This interview is lightly edited for clarity. Sasan Goodarzi, welcome to Stratechery. Sasan Goodarzi: Thank you for having me, I’m a big fan of all the work that you do. It’s great to be a part of it. Well, we’ll get to my relationship with Intuit, which I’ve had for many years. Big fan? I’m going to have to think about that one. We are here to talk about Intuit, a company that you have been at, with a brief break, for over 20 years. Before we get to that, however, tell me about your background, which is a very interesting one. Let’s start with where you were born. SG: Yes, I was born in Tehran, in Iran, and I am the youngest of three — 13 years younger than the other two brothers. I was told I was special, I think I was an accident, so there you go. And back then, you’d ship your kids to the U.S. because everybody wanted to come to the U.S. to study and so my brothers were in the U.S. already studying, and my parents decided to do an experiment with me and shipped me to the U.S. at the age of nine and I was very fortunate, because the revolution actually happened. Got it. So you had already left before then, I wasn’t quite sure on that timing. SG: Yeah, yeah. I was very fortunate, I left before the revolution, about eight to nine months before the revolution, and have been in the U.S. ever since. Where did you end up in the U.S.? SG: I came to Orlando, Florida of all places, because my brothers were going to the University of Central Florida. And, by the way, back then we used to say, “U Can’t Finish”, that’s what we called it, but it’s a great university. But anyways, for that reason, I was in Orlando until my early 20s, until I started being moved by Honeywell. Every couple of years, I was going to a different state, which was a lot of fun, and eventually ended up in California. You studied electrical engineering, right? And then you were at Honeywell and Invensys. What were you doing there? I mean, it doesn’t seem quite the obvious path to being CEO of Intuit, to say the least. SG: Not at all. In fact, it’s advice I give, career advice I give to folks. I actually started at Siemens, I was a technical marketing person, which really — I don’t know why they called it marketing, it was all a technical role that supported our field sales folks. I was in the power generation business unit at Siemens so I used to go pitch why our power generation was better than anybody else’s. I don’t think they need much marketing these days in the Siemens Power Generation Unit. SG: (laughing) Yeah that’s right, they don’t. I did that for four years and frankly, I didn’t know what I wanted to do with my life, because in the meantime, I was doing startups on the side. In fact, before I even started at Siemens, I did a startup called Lazer Cables — with, at that time, a very good friend that was actually the genius behind the idea of fiber optics for the music industry, so I was dabbling in a lot of different stuff. When I was at Siemens, I love the company, but I was wondering, “Is the corporate world cut out for me?”, and I got recruited to Honeywell. And at Honeywell, they put me — it was like business development. In essence, they put me in charge of something they were experimenting with, which was just down my alley, because it was, in essence, a startup as part of Honeywell and I fell in love with the company because of that. I was there, I think like eight years. So is this — was this going to different states? Was that because Honeywell, being a defense contractor, they just have to source from everywhere? Is that the connection? SG: Yeah, but I was in the industrial home automation division. So if you look at my background, I’ve been in power generation, I’ve been in home and buildings automation, I’ve been in industrial, which means I was selling software and hardware at the time to oil companies that need to manage all of their oil and gas in the fields. I’ve been in all of those different industries and then, to your point, I got a call from Honeywell, so my whole background is enterprise and oil and gas and all those different industries. So my forte was getting into something and diagnosing quickly what was going on and what should we do about it, my expertise wasn’t in any one industry. And then, when at the time Intuit called me, I didn’t even know what the company was. Right. It’s a very non-obvious connection, so what did they see? How did they find you? SG: I think the reason they found me is, at the time, the CEO had come from GE Capital, Steve Bennett . And I think at the time, they were looking for folks that knew how to run businesses, not — these are their words, not mine — not a typical Silicon Valley grown-up leader, and that’s what attracted them to me. And I was trying to figure out what Intuit was, because, one, I had never served consumers, businesses — it was always enterprise. But I really fell in love with, at that time, the culture of the company and the leaders of the company. And when I joined, I’ve been here sort of ever since — 20-some-odd years. Well, you did leave briefly to be CEO of energy company Nexant. What advantages do you think you gained from not having at all a traditional Silicon Valley CEO background? And also, on the flip side, what challenges has that presented? SG: The biggest learning I had, which had to come innate because I was in a new job almost every three years, is curiosity — is going in and really diagnosing, really trying to get to the root cause of if something is going well, “Why?”, if it’s not going well, “Why?”. How do you assess the industry, the talent, the priorities? And so I became, only because I had to, a master of how to be curious and diagnose. Does this require a very systemic sort of thinking about the problem? What makes a good generalist? Because you, I mean, clearly you’re a generalist. So what makes that effective? SG: It is, in fact, what you just said. One, you have to be a systems thinker. You have to ask a lot of questions end-to-end, because, one, it’s not from one particular craft, and you don’t know the industry that well and so you have to ask a lot of questions. You have to understand what the customer problems are. Structurally, are you advantaged or disadvantaged? What kind of talent do you have? It requires a systems thinker. But I think the second is you learn how to hire great, great people. Why? Because you don’t know anything about the industry. Once you know what you got to go do, you need great people that know how to go do it, I think those are the advantages. The disadvantages can be that you don’t innately — like when I went from power generation business unit to serving oil and gas, they don’t have too many things in common, and so the disadvantage is intuitively, right off the bat, you wouldn’t know what to do. I think that’s the disadvantage. I would tell you, from what I have learned, I would take those disadvantages any day over a leader that’s curious, that can diagnose really, really well. Because it naturally — one of the things I love about Amazon is the Day 1 mentality. Every day is Day 1, and Day 1 truly is when you’re curious, and you’re trying to understand what’s going on, and what kind of impact can you have. Well, when you don’t know an industry, it is truly Day 1. If you know an industry very, very well, you actually have to work hard to ensure that you’re diagnosing versus, “I have the answers to everything”. I think people would generally think oil and gas, energy, tech, totally different. Is there, though, a commonality in terms of where the investment takes place and the marginal cost of serving customers over time? Is that similar, or are there still really important differences there? SG: There are a lot of similarities I have learned over time in terms of just how do you run a business. So it’s just general management is the commonality? SG: And, by the way, I actually think it applies in this environment more than ever, frankly, because structural advantages — are there any in this environment? What is your structural advantage? Making sure that you’re curious about how you really reinvent experiences because of what’s possible with AI. I think there’s a lot of commonalities, inclusive of — this is probably the biggest thing I would say when I came, and I have a lot of respect, by the way, for all the companies and all the leaders in the Valley because, frankly, we are the ecosystem that drives global innovation, period, end of story. With that said, I think one of the things I’ve learned over my career is we can end up being sloppy in the Valley in terms of how we think about expenses, how we think about cost, how we think about allocation of capital. Those are things you just — when you come from oil, gas, power generation, where margins are thin, room for error is zero because, trust me, you do one thing wrong in oil and gas, you can blow up oil tanks. And so those things carry forward in terms of how do you build quality software, where do you put your money. But then there are things that don’t carry over, which is oil and gas has nothing to do with how consumers consume stuff. Right. Can generalists — could it be a sort of weakness of generalists — you said sloppy. And I think about the generalist angle and systems thinking and it’s easy to skim over the details, “Not that important”. Is that a challenge that you sometimes think about? SG: First of all, I think every leader, no matter what industry, has to have helicopter skills. I think you have to — one, I don’t think you can set clear vision of where you need to go and then move fast to get there if you’re not grounded in details. What does that mean? I mean, for me, details are, are you meeting with customers? Are you in your KPIs? Are you using products? Are you looking around corners in terms of what has to happen in your experiences? I think you have to be detail-oriented, and not just in this environment, I think every senior leader at every level has to have detailed orientation. Systems thinking doesn’t mean you lack detail orientation, it just means that you’re thinking about a problem end-to-end and because one of the biggest things I learned at Honeywell is — think about building a plane, just as an illustrative example, end-to-end, from the moment you say, “Build the plane”, to the moment that plane is on a runway — and thinking about it wing-to-wing. So you can end up building the engine much, much faster, but if you don’t actually think about wing-to-wing — what it takes to build the plane and how to reduce the cycle time of building the plane — then you don’t ship better and faster planes in a shorter period of time. So this notion of wing-to-wing is similar to end-to-end systems thinking. You have to really figure out how to draw the box because when you redraw the box, you redraw your TAM, you redraw what’s possible and that doesn’t mean lacking detail, it just means both become very important. If AI is very good at details, is AI the biggest gift to generalists ever ? SG: I think — by the way, I was on the record, you can go back and look, eight years ago, we went in on data and AI. And the stuff that exists today, of course, did not exist eight years ago and I was on the record saying — and many laughed at me — on the record saying that, next to electricity and the Internet, it would spark the largest global innovation that we could have ever imagined. I think it’s, for many, while it has a lot of risks, it’s going to level the playing field in terms of what’s possible, what you can do, what I can do, what many can do. It’s commoditizing a lot of stuff — and not just information, but insights and helping you make decisions, what it can potentially do to healthcare, what it can do to robotics, what it can do to space. I just think it’s freaking fascinating what’s possible. Is AI a systems thinker or is it a details thinker? SG: Oh, I think, in my view, it’s all about how you design it, that’s the biggest thing we’ve learned in all the agentic work that we’re doing — is it’s only as good as the context you give it, it’s only as good as the data that you give it. The biggest mistake in learning we had in the last couple of years is we, to answer your question, we launched a bunch of AI agents , a payments AI agent, a payroll AI agent, an accounting AI agent, and while they did everything in slices very well, they didn’t actually deliver a great experience end-to-end for customers. That’s because of what the capabilities were about a year ago, but also just how we had designed it, we didn’t design it as a systems thinker. I think it can be both and of course the capabilities are just growing day by day. But it comes down to the context and how you design it, in my view. I did want to do a little bit of a historical tour. It’s interesting to talk to you about Intuit. You’ve been there for 20 years and yet only been there for half of Intuit’s history, this is one of the granddaddies of Silicon Valley. So your first product — Intuit’s, not yours personally, but Intuit’s first product — was Quicken , founded way back in 1983. The company started QuickBooks in 1992, which underpins your largest business today, of which I am a longtime customer. Was Intuit one of the first examples of productivity software companies realizing there’s a lot more money selling to businesses than selling to consumers? SG: First of all, none of us would be here but for Scott Cook , our founder, that was watching his wife, Signe, try to balance the checkbook. I’m old enough, I remember I was taught how to balance the checkbook, I remember doing it way back in the day. SG: We still have that table here, right outside of this building that I’m in, that signifies why Intuit was born. But to answer your question, it was first really the company was invented based on, there’s a better way just to manage your budget, your checkbook, and your financials, which gave birth to Quicken. But really the biggest next “aha” was that businesses were trying to use Quicken to actually manage their business and that’s really the “aha” that Scott Cook and the team had at the time, which is, “Wait a minute, rather than making Quicken better, there is this thing that we can actually invent that helps businesses, manage and be organized and manage their books”. That’s what gave birth to QuickBooks and the QuickBooks franchise, which over time became something incredibly special. Right. Your customers were misusing the product and sort of pulled the new one out of you. SG: That’s right. And Quicken, by the way, ended up only being like a $100 million business, but the brand was so much bigger than $100 million, but it really gave birth to QuickBooks. And then over time — this is a company, by the way, it’s made up of acquisitions. Many don’t know the TurboTax of today was an acquisition of Chipsoft . I have a question about that. On this point to the QuickBooks transition — well, I mean, QuickBooks Online launched in 2000. Again, very, very early to serve something online. The actual transition of your business to the cloud took a lot longer. I’m just curious — you joined, in retrospect, it was the beginning of that transition, but still four or five years after it was online. What do you remember about that transition? What was the challenge there? Was it technological? Was it consumer education? Was it internal, your business model — going from package software to a SaaS model? What was hard about that? SG: I love your question, because it actually starts with where you started, which is, this is a 40-year-old-plus company, and there aren’t too many that have survived. Because remember, we were a company that was born in the era of desktop, right? Yep, package software. SG: We were shipping CDs. No, because we use the word “package software” today, but no one realized — no, it actually used to be packaged. It was a box you went to the store to buy, with — not a CD — with discs inside of it. SG: Thank you! Yes. By the way, that’s actually why I use the words “we used to ship stuff”, we used to ship disks and CDs because many in today’s environment, when you say that, they’re not sure what you mean by “packaged.” I joined right at the beginning of that. And the challenge — I tell you, I give the company and all the leadership prior to me a lot of credit, because there aren’t too many companies that were born in the era of DOS and had to go to the cloud, to mobile, and then the transition to AI. And the hardest part is both cultural — you’ve got a bunch of talent that is building what ends up on a disk, and then you ship the disk. It’s cultural, it’s technology, all of your code is siloed, all of your code is in a bunch of different cans called disks, none of them talk to each other. Yep, and it has to be finished. There’s no updates. SG: And there’s no updates! That’s right. And so it’s cultural, it’s technological, it’s talent, and then there is the dilemma of, “Wait a minute, so you want me to, in TurboTax, build a cloud version, but we’re going to get disrupted, we’re going to disrupt our own desktop business that’s a great business and it’s highly profitable?, so then there is the Innovator’s Dilemma . SG: If you make this shift, how do you make it? So cultural, technological, talent, and then there’s the disruption of your revenue stream. Those are all massive challenges — which, by the way, that’s why we were late to the cloud, because we still have, I think, I’m going to go off of memory, several billion dollars of business that’s desktop. Now, we’re not shipping CDs and disks anymore, it’s sort of still in the cloud. But there are a lot of customers that love the desktop versions, and that’s the biggest challenge, the last one is the customer doesn’t want to change. Yep. Yeah. You mentioned TurboTax earlier, an acquisition, you’ve done a lot of acquisitions over time. I can’t remember — did you acquire TurboTax? I guess it was before you eventually sold Quicken. SG: That’s right, long before I joined. Right, eventually Quicken you sold off ages ago , TurboTax is your consumer play. To my question before — about generally it’s easier to sell productivity software to businesses than to consumers — should there be an asterisk that the way to sell to consumers is if you have the monopoly of violence on your side? Pay your taxes or else! SG: Compliance businesses are great, and that’s really what TurboTax is. That’s the ultimate compliance business. SG: It’s the ultimate compliance business, because it’s all based on the system of taxes in the U.S. and Canada, and you have to do your taxes or you pay a penalty. Now you have to keep inventing, and there’s a lot of competitors in this space, but there’s nothing like compliance businesses, they’re great businesses. Why shouldn’t the government just do the taxes for you? Especially for W-2 workers that are very basic. SG: You have to change the whole system. The government has tried. Right. Intuit has been consistently opposed to it, I think, for understandable reasons. SG: Well, we have been, because, by the way, it’s not free. When the government creates software, you’re paying for it as a taxpayer. Plus, a lot of these W-2 workers, they already have free available to them, whether it’s through us or H&R Block or many other tax prep companies. Unless we change the entire tax system, at the end of the day, free is already available to almost all Americans. By the way, including yourself — if you really want to go do your taxes for free, you could, it’s available to you today. Here’s the question about that — who is the constituency for not changing? Because you guys get cast as the bad guys, “The reason why this isn’t easy is because Intuit lobbies, blah, blah, blah”, which either one, you have the most effective lobbying operation in the history, I mean, you do lobby, I’m not going to deny that’s a fact — but at the end of the day, a lot of people complain about taxes every year and yet it’s still this way. So what are the constituencies that are standing shoulder-to-shoulder with you, as it were, in the system being as it is? SG: There’s two things we’re huge proponents of and, by the way, have been — in my 20 years in the company — when I joined, and wasn’t even running the tax business, these two things were true, and they’re still true today. One, we’re always advocating for simplifying the tax system, because the tax system is way too complicated and the more you simplify, the better it is for the consumer and the business. The other thing that we’ve advocated for is the government shouldn’t invest to build software for taxes, because it already exists and for the majority of Americans, if not all, free already exists. The government spending taxpayers’ money to create software that already exists doesn’t make any sense so those are the two big things that we have advocated for. And the reality is, every year, in some way, shape, or form, with additives of legislation, we unintentionally make the tax system a little bit more complicated. Right. That’s the constituent answer, isn’t it? Everyone is a constituency for their pet part of the tax code, and so it just never changes. SG: That’s right, that’s exactly right. From the consumer perspective, where does Credit Karma fit in this? Which, also by the way, has its own tax product. Is this just a realization, “TurboTax is a very high-margin business”, but at the end of the day, what’s the link? Are you cannibalizing yourself? To what end? SG: Let me maybe just step back and give you the strategic context very, very briefly. So I’ve run all parts of the company before I stepped into this role almost eight years ago in January. And when we declared our strategy, which was, in essence, an AI-driven expert platform with three big bets — which I can briefly get into in a moment — there was a couple of big from-tos. One, I wanted to shift the company from just being a tax and accounting platform to a company that was a platform company that helped businesses from lead to cash and helped consumers from credit building to wealth building, because I felt that, over time, one, just being tax and accounting was too small of an addressable market. And two, every business that I ran — both tax and our business group — the biggest thing I would hear from customers, our TurboTax customers would be, “Why can’t you help me grow my refund?”, “Why can’t you help me, with all the data that you have, connect me to financial products that are right for me?”, “How come you can’t give me immediate access to my refund?”, I would hear a lot of that, and nothing about, “Make your tax product better”. And when I went to the business group, all I would hear is, “I don’t really care about books and accounting, I have to do it”, and my accountant tells me, “I have to do it”, but, what about getting customers? What about managing my cash? What about helping me with whether or not I should buy inventory? So that’s the learning that I had that informed, we got to shift from just tax and accounting to an end-to-end platform company. We invested a lot in building out the platform, but I felt like there were a couple of acquisitions we needed to make and so now, to answer your question around Credit Karma — Credit Karma has like 45 million monthly active users that engage more than five times a month — taxes, once a year. So what I wanted to do is buy the platform and then integrate them to have one consumer platform where you can manage your money, do your taxes, and get access to financial products that are personalized based on your data that are right for you. That was the reason behind Credit Karma, to really create a consumer platform because the hope was, and we’re executing against that, that the more I can engage you throughout the year, the more you’re also compelled, based on everything that I know about you, to get you the biggest refund and for you to do your taxes with us and that was the reason behind the Credit Karma acquisition. Credit Karma’s ace in the hole is, it’s credit, right? People want to know what their credit score is, monitor their credit. Has that always been its biggest benefit, is customer acquisition ? SG: Yeah, that’s the biggest job that it solved, is, “What’s your credit score?”, but then really, when I bought Credit Karma, it was a data and AI platform. We have over 60,000 data points per consumer and based on all the data points, I can connect you — when you need a credit card, to a credit card that’s specifically right for you, versus just sell you a credit card. I’m agnostic, same thing with personal loans, insurance. So now it’s much more than credit score, it’s actually how we connect you to financial products that are right for you, that are personalized, how we help you with getting early access to your refund. It’s credit cards, personal loans, insurance, etc, so it’s now truly — we pretty much do everything for you, except we don’t help you trade stocks. What happens when people just start asking AI for this? Is there an aspect where you finally got to the right consumer business model — it needs to be free, you can get all these affiliate fees by connecting those different services — and then as soon as you finally get that cracked, people are like, “I’ll just ask ChatGPT”. SG: That’s right, this is where we started the conversation — where curiosity and always diagnosing matters so much, so we don’t fall in love with what we’re doing today. Let me answer your question with a perspective and a point of view about the very near future that, listen, may or may not change, but I think it’s important to start there before I answer your question. I believe that we will live in a world where, in some way, shape, or form, agents are doing a lot of the work with you and for you, with your approval and it will make life a lot easier for you. And so whether it’s booking your travel, whether it’s what is the right credit card for you, where you should do your taxes — I believe in a world where agents will do a lot of that work for you and that’s really why we have set out to build a platform where we want to be the financial intelligence layer for consumers, businesses, and accountants. What that means is I want to automate as much of the work for you, and give you intelligence, and carry out the action for you. Because if you think about what we do, we move people’s money, we help them with payroll, we help them with books, we help them with taxes and in the case of our consumer platform, it’s money, taxes, and financial products. That’s why we’ve integrated into a bunch of them — Muse , ChatGPT , Anthropic — because you can ask about, “What credit cards are right for me?”, but ultimately, when you go act on which one is the best one for me, that’s where AI agents, as we do our job right, will call on Credit Karma, will call on TurboTax, to not just give you information. But that’s a tougher spot to be, right? As opposed to most people starting with Credit Karma — where you’re actually the front door, you’re owning the customer relationship — now you’re downstream, you’ve been disintermediated. By the way, I think you’ve had executives say that channel maybe hasn’t been performing as well as you would like, is this a more aspirational goal than a reality? SG: It’s the reality we’re executing. First of all, out of the couple of acquisitions that I’ve done, Credit Karma has really worked out well. Right. I have questions about another one that is also close to me. But we can continue. SG: I am vocal about what works and what doesn’t work, so we can talk about it. But I start there because the whole essence of Credit Karma is really — we’ve executed on the platform strategy, which is create one consumer platform and really own the customer relationship and engagement. That’s the question — you build up this whole platform, and then the whole key component, the acquisition part, what happened? SG: But if you think about the brand that we have, and if you think about the capabilities that we have, we actually have better-intent customers that engage with us now through the LLMs than we did before. Right. So you get fewer customers maybe, but better customers. SG: Better quality and, by the way, not only do they still come to Credit Karma — because Credit Karma is not search-driven, it’s authenticated, it’s an app where we engage the customer. We help you with, “Hey, here’s what’s changing your credit score, here’s a card that’s right for you based on your situation”, so we’re proactively engaging the customer. And if a new customer also wants to ask, “Hey, what’s the best credit card for me”, and this is some of my information, one of the things that we’re getting LLMs to do, first and foremost, is, hey, connect to Credit Karma. Is this the new SEO? Like you need to be the service that the LLM recommends? SG: Yeah, and I think it will be even better. The reality is people now do search in LLMs versus just in Google, but I think one of the things that we’re doing is we’ve built out our experiences and continue to build out our experiences. And this is why, by the way, the OpenAI’s and the Anthropic’s come to us and we have a deep partnership — because we’re executing financial decisions, it’s not just information, it’s people’s money, it’s people’s taxes, and getting it right matters. And so for us, using the word “headless” — being within these LLMs where we get called on as the financial intelligence platform to do the work — is ultimately the way we’re building out our experiences and we think over time, it actually builds out the top of our funnel it actually expands the TAM that we have. Now, that’s yet to be proven over time, but that’s the way we built out our experiences. Presented by I spend a lot of time trying to figure out what is happening to businesses on the internet. What I often find I’m missing is a view across the economy itself. With more than 1.7% of global GDP running on Stripe, working with them means I can bring some of what they’re seeing to Stratechery. They are, in a sense, a barometer of the economy. Take the “SaaSpocalypse”. Earlier this year, nearly a trillion dollars in enterprise software value disappeared amid fears of AI eating its lunch. Stock prices only tell us what investors think will happen, and not necessarily what is happening. In this case, Stripe recently built a SaaS Index using payments data from 72,000 non-AI SaaS businesses. Looking at their actual performance, we can see that while SaaS stocks were selling off, revenue growth was actually accelerating. Year-over-year growth is now over 30%, with some of the fastest growth coming from the youngest SaaS companies. I suspect there are a lot more interesting things hiding in Stripe’s data, and I’m excited about having an excuse to go look for them. We’re going to circle back to a few of those points, but I want to get on the accountant angle. As I mentioned, I’m a longtime QuickBooks customer, the most obvious reason is that is what my accountant wanted me to use. You have a referral program, or you have had one, for accountants. How many of your customers are basically locked in via this channel? I’ve just never really considered anything else, because at the end of the day, I don’t really want to deal with it, and that’s what my accountant wants, is that just the reality for a lot of businesses? SG: By the way, I love where you started the question, because if I could — one minute on what we just rolled out at our Intuit Connect , which is a massive conference for just accountants. And then I’ll answer your question, if I may. Yep, I think it’s very interesting, I was leading to it, so please give me the pitch that you gave just a couple days ago. SG: For sure and we just rolled this out at Intuit Connect with our top accountants there. So, in essence, what we’ve been investing in is a unified financial intelligence platform that serves both sides of the network. One side of the network is accountants, the other side of the network is businesses, and the key is that it’s one unified platform. What does that mean? In shorthand, all the data, all the data models, data ingestion models, all the AI models are all one platform. But the way it shows up for an accountant suite is that the accountant suite helps the accountant manage their firm, manage their work, manage their clients, all in one place and it’s agnostic. If you have QuickBooks customers, other general ledger customers, you can plug it into Intuit Accountant Suite . That’s one side of the network. That’s a pretty significant change, isn’t it, that you’re going to support other ledgers? SG: Significant change, now we’re treating the accountant as a customer, not just as a partner, as a channel. Because in the era of AI, we — and I — had to change our approach to how we innovate for accountants. The other is the business, where not only do we serve the startups and the mid-sized businesses, which is with QuickBooks, but we now have Intuit Enterprise Suite , which is a mid-market suite that helps businesses of several hundred million in size. It’s an AI-native ERP solution that helps a mid-market construction company, wealth management company, architect company, whatever it may be, manage their business. And what we just rolled out at Intuit Connect — I’ll call out three big things. One is Intuit Intelligence , which is, in essence, think Claude or ChatGPT. You can just engage with a chat box if you wish, and you can ask it, if you’re an accountant, you can ask it, “Show me all of my customers and how they compare across profitability and margins for all of my construction businesses, and tell me which ones have their cost out of whack, and how does it compare to their peers”. That’s an example, an illustrative example of our engine doing all the work, but we’re automating the work so we can give you intelligence, which saves accountants. Right now, it’s shifting 30% of their work to advisory versus manual work, because we’re automating a lot of the work. That same capability is available to businesses, where they can just, rather than doing the work across the workflows, automate creating invoices, automate invoice reminders to get paid so a lot of our work around Intuit Intelligence has been around automation and for providing intelligence and insights. The second is we rolled out an enterprise-ready, industry-specific ERP for construction, manufacturing, not-for-profit, which is very disruptive for businesses that outgrow QuickBooks — but now they don’t have to outgrow us, because they’re on Enterprise Suite. Where did they use to go when they outgrew QuickBooks? SG: Typically, they would be either our highest-paying customers and the most pissed off, because they’d outgrow QuickBooks, but they didn’t want to go to the next solution. Right. What would they go to? Like Oracle, or what sort of level? SG: They would go to a NetSuite , as one illustrative example. And for most of these businesses NetSuite is just overkill. And that’s where Intuit Enterprise Suite now really serves an incredibly important sweet spot. And then the last thing we talked about is what I led with, which is Intuit Accountant Suite — now a platform where you can manage your firm, your work, your clients, all in one place and back to your original question of, “I’m using QuickBooks because of my accountant”, well, 65% plus of our customers are “attached” to an accountant, and based on what we’re rolling out, we want that to be 100%, because we see in our numbers the retention of a customer is much higher when they’re using an accountant, because accountants drive success and power their prosperity. Do accountants need you, or do you need accountants? And has that changed? SG: Both, both. I spend 40% of my time with firms — small, medium to large-size firms. And it’s a two-way relationship. And by firms, you mean accounting firms or business firms? SG: Accounting firms. They need us, they need someone — I don’t want to be overconfident — they need someone to help them reinvent themselves in the era of AI. That’s us, that’s what we just rolled out at Intuit Connect, because all of them, if they’re PE-backed, they’re getting a lot of pressure, if they’re not PE-backed, they’re having to compete with firms that are PE-backed. So automation, intelligence, advisory services matters. That’s what we just rolled out. And I need them because it creates a network effect and that really, in the era of AI, is a great place to be when you have a network effect. That’s an actual moat. So is it fair to say your pitch to accountants is basically, “Look, we’re all under the gun here — we’re better off banding together and figuring this out”? SG: Well, if you were to listen to my 25-minute keynote , where I revealed a lot of this product innovation, and you synthesized it and put it in Claude, it would probably tell you exactly what you just said — which is, “You need me, I need you, here’s what the innovation is we’ve done for you and here’s why it’s important”. But does that lead to innovation that actually — is just the reality someone like me is trapped, I’m going to go on with whatever happens and they’re going to figure it out? Or what is the actual threat here, that people just start using AI, the AI companies offer their own services? What are you banding together against? SG: It’s a great question. First of all, if you start with the jobs to be done, and jobs to be solved by businesses — they’re trying to get customers, they’re trying to grow with their customers, they’re trying to understand where their profitability is, they’re trying to get paid, they’re trying to manage their employees. And they’ve got to make sure their accounting and taxes are done right, because clean books. Right. Monopoly on violence — always good to have in your business model. SG: Yes, that’s exactly right. If you start with the jobs to be done, and what’s the role of an accountant? Oh good, that’s one of my questions — what is an accountant’s job? SG: The accountant’s job is to ensure that, one, advisory services — how should you grow your construction company? How should you grow your nonprofit? — making sure their books are clean. Wait, would you have ever said that was an accountant’s job five years ago? SG: No, it’s more important today. And why wouldn’t I have said that? Let’s talk about that. Because I used to run the accounting business at Intuit years ago, this is our pro-tax business, the reality is everything was manual, and all of the time, accountants were spending manually entering data to make sure your books are done right. Now, one of the things we rolled out at Intuit Connect is our agentic books close, where you can upload now your standard operating procedures and processes by firm, because every firm does books differently, and we automate a lot of that. That allows them to spend time on real value-added, “Hey, Sasan, here’s how you should think about growing your business”, “Here’s how many employees you should think about hiring”, “Here’s what inventory you can buy”. Is that a little bit like the mafia helping you? “We’re going to help you close your books”, your books are going to get closed, whether you want them to or not. SG: The reality is, businesses, we see this in our data, and remember, we have 100 million customers, 10 million of them are businesses. Businesses that are connected to accountants, our retention rates and profitability are generally higher. And so this is like real value that accountants provide, and back to your question of what’s different, I think what we’re now doing is making sure that they can run their firm far more effectively by automating and delivering intelligence, so they don’t have to do manual work, so that they can actually do a better job serving the business. And on the other side, we have a platform that the business uses to grow and run their business and it just creates this network effect in ways that I think is beyond, “I need to do manual work and I need to work in the workflow”, to an era of intelligence and insights and how can we grow faster with expected outcomes, where efficiency and profitability must improve because of what’s possible with AI. Are there more or fewer accountants in five years? SG: Fewer, many fewer. And are those accountants actually different people than the accountants today? SG: Well, I think by design, two things in life are true. You die, and you have to do taxes. Right. Is an accountant actually just selling a signature? Is that what they’re actually providing in the future? SG: No, they do a lot more than that, there’s only so much technology can do. Even tomorrow and the day after, as AI advances— SG: First of all, judgment. Closing books — that is really hard. Yeah. By the way, to my listeners, I love accounting. So you’re speaking to the choir here, but you still have to convince me. SG: I didn’t know that, but it’s good that at least this lands with someone that knows accounting. Categorization of all the financial data, making sure that you reconcile everything from a balance sheet to a GL [general ledger], making sure that all the revenues are recognized correctly and in the right categories. There’s a lot we’re doing to automate, but at the end, the judgment of, “Are the books right?”, “Are they clean?”, and then what choices and decisions can you make — that judgment and that intelligence will always, always be important. But I think if you think about a continuum where before it was 90% manual work and 10% judgment and intelligence to eventually, I don’t know when, a year from now, two years from now, it’ll be 90% intelligence and judgment and 10% manual work. I think about it just on a personal level. I had a longtime accountant, who unfortunately retired a couple years ago, and she loved me, she dealt with much larger companies and things. But at the time I was living abroad, I had this very high-cash-flow business. She’s like, “This is so interesting, you present so many weird edge cases and challenges that I’m not used to”, and I almost feel bad for her, because in a world of AI, she would crush it, because she’s exactly the accountant you’re talking about. The question I have, though, is this goes back to you being a generalist. I see someone like you — AI is incredible. I’d be curious to know what, if any, personal AI projects you might be working on because if you’re anything like me, it’s like, this is just so fun — you can do basically anything. That’s not your typical accountant, an accountant is like, there’s the beauty in the numbers, everything has to balance, credits versus debits, and also just a lot of drudgery, actually filling in the forms and making the files correctly, very conscientious, very careful, very high values — all of which seem pretty threatened. That’s why I’m wondering, not just how many accountants will there be, but what does an accountant look like? SG: First of all, there’ll be a lot fewer accountants because of what’s happening structurally, which is PE firms are gobbling them up . Was that happening before AI, or is that a post-AI phenomenon? SG: It was happening beforehand, because there’s just a lot of inefficiency, and PE firms look at that and say, “There’s a lot we can do”, and it’s just accelerating now with what’s possible with AI. I think that’s one reason. I think the other is I believe there will be accountants that are all-in on AI and accountants that are not and those that are not all-in on AI are the ones that are going to be disrupted and won’t be around two to three years from now. So that’s why, in general, I feel like there’s going to be fewer and fewer accountants. But here’s what I would tell you — and listen, you’re an accountant, I spend 40% of my time with them. I’m not an accountant, I just appreciate it. SG: The biggest thing I hear from them is they don’t really like doing debits and credits, they don’t like doing the manual work, and it’s the biggest cultural change they are focused on driving. And a big part of that, that we’re helping them with, is either training their folks, and/or for them to upgrade. I do think there’s going to be a massive change and upgrade of talent, either those that love the manual work are going to realize, “I have to love automation”, not doing manual work, or they’re just going to go get somebody else that will love that work — particularly these folks that are coming right out of college and want to be accountants and want to be CPAs, they’re much more tech-savvy, and they will want to drive automation. I think that’s the change you’re going to ultimately see, and we’re seeing it right now, because the top partners are driving it. If you look at your recent financial results, you’ve been growing revenue amongst your core customers, but barely growing total customers. Now you’re talking about going up-market, like into the accounting workstream, for example. You mentioned with businesses, your recent focus on ERP services. Again, going up-market, between that gap between NetSuite and yourself. All that makes sense. At the same time, you’ve brought up disruption, jobs to be done, you’re a student of this theory. A company going up-market, increasing average revenue per user, even as there’s a technological shift at the bottom that is gobbling up the business there — that sounds like a company that’s being disrupted ? SG: I love your question, I’m glad you’re asking it now versus four years ago. So we declared, in essence, these three bets that I’m talking about eight years ago. And what are those bets? If I were to simplify them, one is to disrupt assisted tax — out of the total tax TAM of $40 billion, $35 billion is assisted taxes — when you go have somebody else do your taxes for you, a human doing it. And it’s the signature that matters. Is the goal there that customers no longer need to hire accountants? SG: I’ll come back to that. So that’s one space. The second big space was shift from tax and accounting only to money, our money benefits and money portfolio being the core of our platform. What does that mean? It means bill pay, it means payments, it means instant deposit. And then the third is go to mid-market. When I declared this nearly seven-and-a-half, eight years ago, investors were very skeptical of me, because what they said was, “These are markets you don’t play in, you could screw up your margins, I’m very skeptical”. If you now fast-forward to where we are — to answer your question — it’s now 30% of the company, so it’s like $7 billion-plus, growing at 30%, we’ve already proven these. So is the answer to the old business is, we’re just going to outrun it with the new business? SG: That’s right. So AI is a huge tailwind for us, because we’ve been investing in it in these areas. The core business is — if you go back to Investor Day and what I just shared, I said, hey, we’re growing 30% of the company, my core business has slowed. I need to increase my market share, so I’m changing my lineup. Right, because isn’t the problem that your core business — maybe it’s not very sexy — but that’s the foundation of your new businesses? SG: That’s exactly right. And so that’s why we, in essence, what we’re doing is reinventing our product, but also being disruptive on price to take market share, while at the same time scaling my big bets because I want to make sure five years from now, we’re a healthy and growing company. Because here’s what I believe will happen with AI in our industry and in the spaces that I play in, AI is a tailwind for me with these big bets — disrupting mid-market, disrupting assisted [tax] — Just because you can develop software more quickly, or what? SG: My biggest differentiation in assisted tax is I have scaled the one thing that matters most, which is the human signature, virtually, at scale and I have a disruptive price. So I can do anybody’s taxes for them at a disruptive price with an amazing virtual experience, where you never have to leave your place of work or home and it’s an amazing experience. The differentiation is I have scaled the human signature because of AI. So that’s an example of — and my competitors are a lot of mom-and-pop shops and so scaling that for the next 10 years. You were saying “the signature”, are you talking about actual accountant signatures, or are you talking about, like, the TurboTax product where you’re actually— SG: We have about 18,000 experts that are CPAs, accountants, on our platform and those are the ones that are virtually doing people’s taxes for them. Got it, got it. So that’s a separate pool than your accountant business, I was conflating them in my head. SG: That’s right. I mean, the accountants that we’ve been talking about for the last 45 minutes are accountants that, they don’t even want to mess with somebody’s tax return unless it’s greater than $6,000, we don’t compete with them, we’re helping them grow their business. These are much more the normal Americans and the taxes that they pay. The reason I think we got on this topic is, for me, that’s a huge tailwind, AI is a tailwind to disrupt those industries. But I think for the lower end of the market, AI can help you reinvent experiences but if you’re not priced right, you’re going to get disrupted and that’s really what I’ve done. In the low end of the market, I’m reinventing experiences, but I’m adjusting my price-to-value ratio, so I can win these customers when they first come into the market, and then I can grow with them over time. I do have to ask you, because I hinted at it before — you acquired Mailchimp for $12 billion in 2021, a company that I originally worked with and left in part because I feel they send email people don’t want, and I send email people do want. The idea was extending QuickBooks from accounting into customer acquisition, that platform stopped growing, you’re segmenting it off , is that an acquisition that didn’t work out? SG: First of all, legendary brand, legendary product, legendary employees — full stop. I’ll come back to that in a moment. What we want to do hasn’t changed, the customer problem of helping customers grow has not changed, and so now we’ve built that AI natively within QuickBooks. Right, but you said before you’re building a money platform. Is the issue that Mailchimp doesn’t have anything to do with money, at least directly? It’s a secondary input. SG: Growing customers has everything to do with money, because eventually, they have to pay. So when I say “money”, it is, in fact, owning the right to the workflows that lead to money — which is why solving the problem of getting customers, which is what Mailchimp does, is something we’ve built out AI-natively within QuickBooks. But to your point, we have carved out — I don’t think we’ve executed well on Mailchimp, so we’ve carved it out separately to really focus on getting the business right and making sure it’s transparent how we’re performing externally and as I’ve said to investors and our employees, all options are on the table to make sure that it has the right future. All right, I want to get back to the more interesting question, because we’re almost at time here. To go back, to put this together historically — you went through one big business model change, from package software to subscriptions. What does the business model look like in 10 years? People talk about usage-based, outcome-based pricing — you’re not necessarily seat-based , so that’s a big advantage for you, you also have the affiliate marketing angle on your consumer side. What’s the core way that Intuit makes money in the long run, and is that going to be fundamentally different? SG: Three things I will say. One, first of all, you’re absolutely right — we’re not seat-based, so let’s get that foundation confirmed by me, that we’re not seat-based. I’ll say three things. The way we’re set up today is, we’re beyond user-paid, so the entire Credit Karma platform — everything for the user is free, but we get paid based on all of our partners on our platform, so that’s one business model. The other business model is subscription-based, which is our business platform is subscription-based — about 80%, 90% of it. The third is we’re transaction-based. Right. You file your taxes, you pay a certain amount of money. SG: That’s right. Or you invoice, and we get paid, a customer gets paid, that’s transactional. So those are the three models in which we operate today. If I were to answer your question five years from now, three years from now, not just 10 years from now, I think what will get added to that is consumption-based. SG: For instance, a lot of what we just rolled out at Intuit Connect is, if an accountant uses certain agentic capabilities, like creating specific tailored experiences for a construction company, and that construction company pays the accountant more for those services, which they will — we have the opportunity to monetize that consumption. Is that monetization — is that usage-based, or do you envision outcome-based pricing ? I mean, you are dealing with money. SG: No, no, it’ll be usage-based. Is there room for an outcome-based pricing model in any of this? SG: You know, there could be — I don’t want to fixate on “yes or no”, we test a lot of this with customers. What I would tell you — customers today, doesn’t mean tomorrow won’t change — generally, customers aren’t that interested in outcome-based pricing. They’re generally not interested in token-based pricing, because then they can’t control what they’re going to end up paying, so they still want an element of fixation of what I’m going to pay. Predictability, yeah. SG: That’s right. I think three years from now, the value of what they’re willing to pay for in certain markets will go up. In certain markets, it could go down. But I believe in three years, we’ll still have beyond user-paid, we’ll still have transactional, we’ll still have subscription, and I think there’s going to be a smaller element — and growing — that will be consumption-based. Do you feel like you’re on offense or defense? SG: Every day, we’re focused on playing offense and every day, I open my email to see what was launched yesterday, to figure out where we need to play defense. Very good. Sasan Goodarzi, thanks for coming on to Stratechery. SG: All right. Thank you for having me. This Daily Update Interview is also available as a podcast. To receive it in your podcast player, visit Stratechery . The Daily Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly. Thanks for being a supporter, and have a great day!

0 views
Stratechery 1 weeks ago

AI on the Ballot

Listen to this post : From Bloomberg this morning at 2 a.m. Central: Democrats won control of both chambers of the US Congress in a midterm election landslide that amounted to a rejection of President Donald Trump following widespread voter discontent with the economy and the war in Iran. Across the country, Democrats scored victories in GOP strongholds, flipping at least seven Senate seats and winning House races in districts Republicans easily carried in 2024. Several news organizations projected that Democrats would win the House, while Decision Desk HQ projected they took the Senate. I have no doubt that Iran and the economy played significant roles in this election; what stood out to me, however, was very much Stratechery fodder. Specifically, I watch a lot of sports, and particularly this fall, watched a lot of baseball. My Milwaukee Brewers — excuse me, my World Champion Milwaukee Brewers — demanded my full attention, which also meant I saw a lot of campaign ads, and by far the most common topic, at least here in Wisconsin, was data centers. Moreover, it was never entirely clear which way it cut politically: both parties were anti-data center, accusing their opponent of being the opposite. Meanwhile in Texas, from Yahoo Finance : A strong showing by Democrats in Texas swept Gina Hinojosa into office as the state’s first Democratic governor since 1990 in a contest closely watched by Big Tech companies. The state is considered a bellwether for government restrictions on AI data centers…Hinojosa, Texas’ first Latina governor, has pledged to impose a moratorium on data center construction in Texas, which accounts for more than 25% of planned US capacity by 2030. Republican incumbent Greg Abbott, I would note, already declared a moratorium on new data-center permits in September; that’s how much of a hot-button issue this is, including in the state that is the biggest opportunity and beneficiary of the data center buildout. So yeah, this is a Stratechery topic of concern. Earlier this year, while writing about data center opposition , I said: There really wasn’t widespread opposition to data centers when they were being built for the Internet generally; indeed, local governments would compete for them, eager for the economic benefits that came during their construction and the long-term tax benefits once they were operational. In other words, it seems pretty clear that the current wave of opposition can’t be divorced from AI — you know, the technology that in the best case scenario is going to take everyone’s jobs, and in the worst case scenario end the human race. I do not, I don’t think, need to explain to you why this is not a very compelling message, and, in the very long run, I do think that humans will have economic value . At the same time, it’s hard to explain the huge amount of investment that is going into AI, not just at the model level but also in terms of data centers, without admitting that a big part of the thesis is that AI can not just increase the top line for companies, but also increase the bottom line — i.e. replace the jobs of some number of humans. Again, I’m generally an optimist here, but this already happened with the first wave of IT . That leads to an interesting contrast to globalization: when companies were closing down American factories and laying off workers and moving operations to China, none of the affected towns or workers had a say. They just suddenly no longer had a job, and a huge number of cities across the Rust Belt no longer had a reason to exist. People simply had to move, or worse, retreat to things like alcohol or drugs. AI, however, is the opposite: building data centers requires permission, which is to say that people actually have a say. Again, I am not at all saying that these people are well informed about data centers, or about the economic impact on their communities, much less the economic impact of AI generally; what I am noting is that people who didn’t have a say in globalization are suddenly finding they do have a say about AI, and it’s not a surprise they are expressing their disapproval by blocking data centers. That message was loud and clear last spring; now it is deafening, and it’s tempting to say that tech, particularly the frontier labs, have no one to blame but themselves. In fact, however, some of the biggest culprits may, in the long run, be some of the biggest beneficiaries of the anti-data center wave. My thesis — driven not just by national results and the Texas governorship, but also the fact that, as this thread documents , nearly every relevant local ballot measure opposing data centers won or is winning ( the only exception so far was in North Dakota ) — is that the constraint on AI dispersion has shifted once again. First we were short on compute, but that ramped; now we are short on memory, but that will ramp; we were coming close to being short on power, but that will be rectified. The final constraint is political, and that may be the most imposing constraint of all. Not only is it harder to solve, but it actually interferes with solving the other constraints. Many of the theories about AI assume that it becomes a commodity; I think it’s just as likely we are looking at scarcity for much longer than we expected, which will have ripple effects on everything. To that end, here — with the caveat that nothing leads to hot takes you regret like an election — are ten perspectives on what a permanently impaired ability to build might mean. To be clear, I’m not declaring permanency — look at how our elections swing back and forth — but it’s worth considering what the implications might be if Americans don’t change their minds about data centers. The most obvious reason why the stock market has been rocky over the last few weeks is the tightening of credit markets, which I wrote about this summer . We are, needless to say, full on into the equity funding era (and while I’m here, I haven’t written about Anthropic’s updated S-1 , but it strongly affirms my initial analysis ). At the same time, it seems clear that this outcome is very bad for the AI trade in the short term, but much of the pain may have already been endured. The Neoclouds have been at the forefront of credit concerns, and from a certain perspective, being unable to build data centers has a silver lining in that they couldn’t get the money to pay for them anyways. Moreover, Neoclouds with facilities that are already permitted now have an asset that is going to be scarcer than expected. At the same time, it’s Neoclouds that need new sites and new permits, even as they are likely to have much less political sway than the hyperscalers — who already have more sites they can repurpose or rebuild. It was already getting harder to get capital, relative to the hyperscalers; now you have to layer on that it’s going to be harder to get permission, too. The downside for the neoclouds is upside for the hyperscalers, who are more likely to grow their share of AI compute than they were before. Moreover, investors may be pleased if the hyperscalers’ ambitions are somewhat curtailed; political opposition is an excuse to moderate spending, knowing that everyone else is in the same boat. Nvidia will almost certainly be a short-term loser. Fewer data centers mean fewer Nvidia chips generally, and a weakened Neocloud set is particularly harmful not only because Neoclouds almost exclusively use Nvidia full-rack systems, but also because Nvidia is back-stopping a huge number of them. It’s the hyperscalers, meanwhile, who are Nvidia’s biggest long-term threat, given their scale and ability to invest in their own chips. At the same time, a constrained world makes Nvidia’s offerings more compelling. CEO Jensen Huang actually spent an entire keynote making this point back in 2025; I wrote in an Update : When Huang says “iso power” he means that power is the constraint; the question is how many tokens can you generate given a fixed power envelope, and Blackwell can generate a lot more. These were the numbers Huang gave for a 100 MW data center: What this means is that the first order of business for an ASIC maker is to match that level of efficiency. That, though, means keeping up with Nvidia’s product development cadence; The Information has a post about Amazon offering instances using its latest Trainium ASIC at a significant discount to the H100. Huang, though, spent this keynote insulting the H100! You can almost hear him speaking to Amazon directly: you spent all this money, and are offering these huge discounts, just to entice developers to lock themselves into an inferior chip, even as reasoning models are going to swamp your available power generation… Huang is issuing a challenge to every cloud provider tempted to make an ASIC: do you really think that you can keep up with this, and risk running out of power capacity because your chips aren’t efficient enough? Don’t even try. A world without power or data center constraints is one where it is, in the very long run, very difficult to maintain Nvidia’s margins; once those constraints start to bite, however, an advantage in tokens/megawatt is worth a lot of money. Go back to my point about the hyperscalers: one way to generate more tokens is to build a new data center; another way is to rebuild a data center with new Nvidia chips. China is the inverse of the scenario that Huang laid out: the country has plenty of power and the ability to build a lot more, but they don’t have leading-edge chips. Some of the advantages of leading-edge chips, however, can be overcome by simply throwing tons of electricity at the problem. From a 2025 Update about the Huawei CloudMatrix 384 supercomputer: The CloudMatrix 384, which is a supercomputer consisting of 384 chips, takes up dramatically more space, uses dramatically more power, is almost certainly dramatically more expensive to make, and is dramatically more brittle than Nvidia’s NVL72; there is no one in the world who would choose the CloudMatrix 384 over the NVL72 if they had the choice. That’s the thing, though: China doesn’t have the choice, because of U.S. export controls. The country’s saving grace is that they are operating on different constraints than the U.S.…China — which actually builds power generation — is not iso power; they are iso compute: Huawei’s goal with the CloudMatrix 384 is to get as much compute as possible, power efficiency be damned. In the fullness of time — i.e. as China figures out leading-edge chips — this may be an insurmountable advantage. When SpaceX released their S-1, I explicitly pointed to political opposition to data centers as one of the company’s most significant upside scenarios: The third question about data centers in space is if there is a stick. Specifically, while I think that racks-in-space are both a lot more viable than people think, and a lot more relevant to agentic inference than current modes of compute, it is at the end of the day cheaper and easier to build on earth, all things being equal. All things are not equal, however: right now we are at the very beginning of the AI buildout and already one of the biggest constraints is not just power (expected), but zoning (unexpected). I wrote in an Update last week: I already posted the relevant excerpt above. Continuing: In that Update I made the case that data center builders — and by extension the companies that use them — should straight up pay people for permission to build data centers in their communities. At a minimum, however, that increases the costs of terrestrial data centers. What seems very plausible in the long run is that the demand for compute ends up being so large that there eventually is nowhere left to build, making the vast expanses of space not just an alternative but in fact the only choice. This is where I come back to my experience watching these campaign ads: this isn’t a partisan issue, where the pendulum will swing back in two years. Both parties see opposition to data centers as a winning cause, and not just nationally: that thread has multiple examples of local officials straight up recalled for approving data centers in their localities. Space will be incredibly difficult, but difficulty is all relative! Moreover, to the extent opposition is permanent, the more this actually reduces the risk of data centers in space, because investors will have more confidence in there being enduring demand. If the hyperscalers had a red queen problem , the frontier labs have it even worse: the biggest revelation from Anthropic’s S-1 was that their inference margins were very high (good!), which meant that their massive compute costs were actually in R&D; this is bad because that R&D spend is governed by competition with the other frontier labs, and absent constraints, that spend looked set to surpass revenue sooner rather than later. Now, however, it is the public that is effectively demanding that the frontier be paced, which is great for the frontier labs’ financial outlook. The upside goes further than that: the point about efficiency being more important in a world of constraints doesn’t just matter for chips, but also for models. Devoting limited compute to inefficient and costlier-to-serve Chinese models is a poor trade if there isn’t enough compute. Moreover, open models may be at greater regulatory risk, as Democrats seem much more likely to pass laws that ultimately favor the frontier labs. There are some counter-arguments. First, if compute is constrained, and thus more expensive, then that will increase the drive to right-size models for jobs, which is a big argument in favor of open models; the fact that compute won’t be a commodity means that the frontier labs’ lower cost to serve will never come to bear on the low end of the market. And, it’s worth noting, it will be harder for the frontier labs to build their own compute, leaving them dependent on the hyperscalers; big tech wins again. The most obvious advantage for Apple is that in a world of constrained compute, inference on the edge will be that much more important. Another advantage is that any slowdown in the dispersion of AI is good for a company that dominated the previous paradigm . The biggest advantage of all, however, is that any slowdown in the AI buildout offers hope that the memory crunch will ease. This will take a while, to be sure, but no company would be a bigger beneficiary of the world running out of places to put AI chips than Apple. AI has drawn a great many battle lines, not just in terms of data centers, but also within AI, and between AI and the rest of tech. Specifically, anything that strengthens the frontier labs is likely bad for the rest of the startup ecosystem. In a world with fewer data centers, startups will have less access to cheap compute, even as a slowdown in capabilities both means it takes longer to build compelling products and also gives the frontier labs time to actually build products that secure their ownership of the end user. One major fear that investors have had about the data center buildout is that there would be too much capacity, causing prices to plummet and investors and creditors to lose their shirts; that scenario, however, would be great for users, who would suddenly have access to incredibly cheap compute. Of course most users aren’t going to use AI directly; what they will miss out on is the equivalent of the Web 2.0 wave of free Internet services that were predicated on compute being free. Now there’s a good chance that AI is going to have meaningful marginal costs for a long time to come. At the same time, while I have been an optimist about humans in an AI future , there is no question that we were and are facing a wrenching change as intelligence becomes a commodity. It follows, however, that if it doesn’t become a commodity — or if that change takes longer than expected — that humanity will have more time to adapt. Maybe this is a break we needed; what is increasingly clear is it’s the break Americans wanted. That noted, there is a long history of citizens demanding the ends and opposing the means, and it’s the demand that wins in the end. The big difference today is that, unlike the first few waves of tech, the industry is going to need to do some convincing, and break the habit of assuming everything it does will get the benefit of the doubt. AI wasn’t technically on the ballot, but it was implicitly, and will be forevermore. Presented by I spend a lot of time trying to figure out what is happening to businesses on the internet. What I often find I’m missing is a view across the economy itself. With more than 1.7% of global GDP running on Stripe, working with them means I can bring some of what they’re seeing to Stratechery. They are, in a sense, a barometer of the economy. Take the “SaaSpocalypse”. Earlier this year, nearly a trillion dollars in enterprise software value disappeared amid fears of AI eating its lunch. Stock prices only tell us what investors think will happen, and not necessarily what is happening. In this case, Stripe recently built a SaaS Index using payments data from 72,000 non-AI SaaS businesses. Looking at their actual performance, we can see that while SaaS stocks were selling off, revenue growth was actually accelerating. Year-over-year growth is now over 30%, with some of the fastest growth coming from the youngest SaaS companies. I suspect there are a lot more interesting things hiding in Stripe’s data, and I’m excited about having an excuse to go look for them.

0 views
Stratechery 1 weeks ago

Apple Earnings, Memory Problems, The Hardware Company Regression

Apple's earnings suggest that the company's hardware business is doing better than ever, even in the face of memory challenges. Does the software story help or hurt?

0 views
Stratechery 2 weeks ago

Google Earnings, Facebook Déjà Vu, The Agent Play

Google's earnings miss looks a lot like Facebook's, and that's a credit to Google. The company is being more aggressive than they get credit for.

0 views
Stratechery 2 weeks ago

2026.44: At Long Last…

Welcome back to This Week in Stratechery! As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone . Additionally, you have complete control over what we send to you. If you don’t want to receive This Week in Stratechery emails (there is no podcast), please uncheck the box in your delivery settings . On that note, here were a few of our favorites this week. Presented by When I write about tech companies, I am often writing about businesses that depend on Stripe , because Stripe provides the foundation for Internet-native companies — including Stratechery. Consider Substack: building Stratechery was complicated and involved gluing together disparate pieces; Substack made it easy by delivering an integrated offering. One piece that was the same, however, was Stripe: Substack creators are also Stripe customers, as are creators on many other platforms. Zoom out and the pattern repeats. Fintechs can build entire business models around card programs without ever having to build the issuing and processing infrastructure underneath it. AI companies can charge for what customers actually use, which is particularly important when usage also drives costs, without having to build the systems to track and bill for it themselves. No business succeeds because of Stripe. Stripe doesn’t answer the hard questions about what to build or whether anyone will actually use it. But it does make more things practical to try. This week’s Stratechery video is on Do You Trust Apple? . Finally, Some Anthropic Numbers.  Monday morning brought us Anthropic’s S-1, and not a moment too soon. After four years of arguments about the economics at frontier labs, we now have some real numbers to sink our teeth into, and Ben did just that with his Update on Tuesday . Come for that analysis, and then head over to this week’s Sharp Tech where we talked in depth about what Anthropic is telling investors, questions that remain unanswered, plus why the SEC forced Anthropic to help us understand Meta’s Muse bet . — Andrew Sharp The End of the Internet as We Know It (and I Feel Fine) . Arriving on the heels of a handful of ominous cybersecurity incidents and with continued upheaval guaranteed, Monday’s Article made two distinct predictions . First, the pressure applied by AI will eventually (or imminently?) rupture the global internet that we’ve come to know for the past 25 years. But second, as society moves to more localized solutions and products, that may in fact be a welcome development for all of us who have come to accept degraded experiences throughout our technological lives. As an avid group chatter who’s more or less abandoned public social media, I can certainly get behind the second take!  — AS Autocracies Then and Now. On Sharp China this week, we began with an extended look at the climate of fear across the Chinese bureaucracy as Xi Jinping continues an anti-corruption crackdown that was absolutely necessary but may be going further than Xi’s earlier proponents ever expected. Then, we turn to a glorious mess surrounding Mao Zedong and international soccer star Kylian Mbappé. On both that episode and in this week’s Sharp Text , I make the point that connipitions like this one, anchored in a bizarre debate over Mao’s governing history, are evidence that rumors of a Chinese century have been greatly exaggerated.  — AS The End of the Internet — The old Internet, where anyone could host a server is over, but it’s just the final blow in an ending that has been coming for years. The future, though, is brighter than it appears. Anthropic S-1, The Cost of Compute, Who Is Paying? — Anthropic’s IPO is finally here; the core business is very profitable, but competition may eat all of the profits, even as investors are set up for endless dilution. Meta Earnings, Muse and the Frontier, Musebook — Meta’s earnings come in the context of Zuckerberg delivering on his promise; if investors believe in Muse they must support Meta’s spending on compute. An Interview with Palantir CTO Shyam Sankar About the AI Religious War — An interview with Palantir CTO Shyam Sankar about AI diffusion and safety, forward-deployed engineer, and what the battle in tech is actually about. Kylian Mbappés Little Red Card — Surveying the biggest scandal in China this week, and another reminder that the CCP cannot get out its own way. The End of the Internet Touchscreen MacBook Pros How Linux (and Windows) Ate UNIX Silicon Valley Walks the Jagged Edge Five Questions on the Fifth Plenum; Mbappé Hysteria and a Chinese Own Goal; The STAR 50 AI Bubble Has Burst The Opening Week in Review, Nuggets Darkness and Eastern Conference Angst, A True Catastrophe for OKC Anthropic Opens the Kimono, Bubble Questions and Frontier Costs, Q&A on Meta, Starlink, Brewers

0 views
Stratechery 2 weeks ago

Sharp Tech on Video: Anthropic’s S-1 and What It Means for Meta and Muse, You Can See Everything, and Brewers!

Sharp Tech is, for the first time, available on video. You have three ways to consume it: YouTube, Spotify, and here on Stratechery.

0 views
Stratechery 2 weeks ago

An Interview with Palantir CTO Shyam Sankar About the AI Religious War

Good morning, Today’s Stratechery Interview is with Palantir CTO Shyam Sankar . Sankar has worked at Palantir for 20 years, and has taken on an increasingly visible role as public intellectual, including authoring the book Mobilize , an exhortation “To Reboot the American Industrial Base and Stop World War III”. I previously interviewed Sankar in June 2023 . That interview is a great place to start in terms of understanding Palantir, which we mostly discuss in this interview in the context of AI. To that end, in this interview we discuss why Sankar thinks the industry is insufficiently focused on AI diffusion, why companies focused on the top-line have more success adopting AI, and why he is frustrated at the labs refusal to take a systems engineering approach to AI safety. Then we discuss what Palantir has learned from AI, how companies can preserve alpha instead of the median, and whether or not we need more models. Sankar also explains exactly what a forward-deployed engineer is, why tech is in a religious war, and his retrospective on Anthropic’s battle with the U.S. government. This Interview is presented by Stripe, the economic infrastructure of AI, and for Stratechery itself. As a reminder, all Stratechery content, including interviews, is available as a podcast; click the link at the top of this email to add Stratechery to your podcast player. On to the Interview: This interview is lightly edited for clarity. Shyam Sankar, welcome back to Stratechery. SS: Ben, thanks for having me back. Well, the last time we talked was in conjunction with Ted Mabrey , the head of Palantir Commercial. As is usually the case with dual interviews, we got right down to business — and I do have to say, I do a lot of interviews where I usually kind of know where I want to go, but that one was super helpful to me in understanding Palantir and thinking about AI generally . But before we get into that, we never got to do my traditional background questions — where you came from, how you got to where you are. So for you, do we start in Orlando, in Lagos, or in India? SS: At least Lagos. My father, in his early 20s, moved to Lagos and built one of the first pharmaceutical manufacturing facilities in all of Africa. He set up this whole life there for the better part of ten years. I was born in India, but really I grew up until I was three in Lagos. And then there was this horrible armed robbery that happened at the house — they decapitated the dog, pistol-whipped my father, and threatened my mother. So we left all of our earthly possessions behind and started over again in Orlando, which ended up being a truly phenomenal place to grow up. The profound optimism of Disney World, the shadow of the Space Coast , seeing shuttle launches in elementary school, running into your courtyard filled with this sense of optimism about how technology will inevitably lead to a better world — which, if you fast-forward twenty years from the ’80s, I think kids were fed a very different diet of Terminator and “ nature always finds a way “, “How dare man try to do anything that creates progress?”, so that’s really where it started. Is that actually the ultimate example of man overcoming nature — not the space shuttle, but in fact Disney World? SS: (laughing) That’s true! The fact that millions of people go there every year, to the middle of the swamp — and it’s fantastic, a lot of fun. What does meritocracy mean to you? I was reading the recent profile of you in Colossus , and it opened with [Palantir CEO] Alex Karp’s statements in that regard, but it was put in the context of you growing up in Orlando, taking an entrance exam to go to Trinity High School, and how that launched you into what you became. Is that a sentiment from Alex, is it from you, or what’s your take? SS: It’s a sentiment for both of us, for different reasons. We’re both kind of acquired tastes, in many ways, and living in a world that judges you by the fruits you bear, rather than any dimension you have no control over, is obviously a better world to live in. I think it’s a precondition to the American Dream. I was going to ask you that. Because you have this background — born in India, originally living in Lagos, come to America. What does being an American mean to you? Is that the same question, or is there some tension between those two things? SS: They overlap profoundly. Being an American is this profound belief in your own human agency, and the community you’re in, and the ability to create a more perfect union — the recognition, the ability to hold the contradiction, that not everything is perfect today but it’s incumbent on us to make it better. That may sound a little Pollyannish and even corny, but that doesn’t exist anywhere else. Everywhere else, people are profoundly cynical — things are the way they are — and there’s a very thin line between cynicism and nihilism, and you kind of give up on everything. That’s the death spiral, that’s how you get the Dark Ages. After university and graduate school, you worked on Xoom, on remittances — and then you go to work for Palantir. SS: Xoom with an X. Xoom with an X, that’s right. Not the Zoom with a Z — which maybe I shouldn’t be talking to, since you’re now an executive at Palantir using Zoom with a Z, but that’s another discussion. You worked on Xoom and remittances, then Palantir, working on national security, arguing for building our homegrown industrial base. There’s an argument there that — to take a paradox that’s maybe unfair — how do you go from a career focused on sending money out of America to being an advocate for money here in America, rebuilding our capital base? Is there tension there, or am I creating paradoxes out of thin air to build a theme? SS: I think you’re creating a bit of a paradox out of thin air. First of all, there’s a policy question — if you have a working class trying to get money to support their family, how can they do that as efficiently as possible? The bigger issue is the profound lie of globalization. The calculus that globalization got wrong is this idea that we can do the innovation and “they” can do the production — “they” make the things, we do the innovation. The problem is that innovation is a consequence of productivity, if you stop making the thing, you cede the stimulus required to do the innovation around the thing. The example I always give young computer scientists is, what do you think motivated Google to do the research behind “ Attention Is All You Need “? If you talk to that team, they wanted the 3% improvement in Google Translate. You cannot think of something more prosaic leading to something more profound. So we should be very careful where we give up — we buy a little too much into comparative advantage and give up the stimulus for production. Why do you think — what leads to that mistake of thinking that it’s okay to just own the inputs and not worry about the outputs? SS: Because at the margin, it’s true. At the margin, you can hire WuXi as a cheap set of hands at the lab bench to do a little bit of work. But at the limit, that’s a transfer of knowledge. At every increment it seems to make sense — you’re still doing the innovation — until one day you wake up and realize, “Oh, I don’t have it anymore”. You’re the only Stratechery interview that I’ve offered a job to. You sent over a list of questions you wanted me to talk about, and I have to say they’re really good — I did ask if you wanted to prepare all my interviews going forward, I don’t think I got a response, maybe the email got lost. SS: (laughing) I’m thinking about it. Okay, good. Well, let’s drill in — we’ll take your outline, why not, go through it and talk about it, and I think there are themes that might emerge. So number one, and this is from Shyam Sankar, does this count as a Shyam bomb, by the way, the long extended email detailing what we should talk about? SS: It’s not nearly vitriolic enough, so you’re spared. Got it, that makes sense. Number one, is the AI industry solving the wrong bottleneck? You argue we are over-investing intellectually in the supply of intelligence and under-investing in diffusion. Tell me what you mean. SS: If you thought about a crass metaphor of AI through electricity — there are people who generate the electrons, the power producers — for electricity to have any impact on our economy, there needed to be other people who built machine tools that run on those electrons. And that led to a profound revolution. Not only would you introduce one piece of machinery into your factory, you’d start redesigning your factories around the fact that we have electrons. Let’s call that diffusion, how does the miracle of electricity diffuse into the economy? There’s a similar question with AI. I think we are very focused on how to make better models, smarter models, more machine intelligence — and we should keep doing that — but we’re very under-indexed on how to accelerate the rate of diffusion. Where is the economic value being created? How do we turn these tokens into economic value more quickly with a great ratio around it? I’m obsessed with that, because I think there’s a license to operate in society where the benefits of this stuff diffusing really do matter and I think we’re getting the calculus around some of this pretty wrong. So who is “we”? Who are the ones insufficiently focused on diffusion? Who are you trying to convince here? SS: I think the labs are, but in general the entire market structure. People are being rewarded — and it’s very rational, if you’re a lab, you’ve been in your own reinforcement-learning loop, you do fundamental research, you make a more powerful model, and your valuation doubles or quadruples and every year it seems to keep working. Baked into that is the idea that eventually this stuff will diffuse, eventually there will be economic value. But one of the conceits I’m not sure people are getting right is, what is the atomic unit of the enterprise? Let’s leave consumer out of this. The presumption, as expressed through the product strategy you see from the labs, is that the atomic unit of the enterprise is the knowledge worker — if I can make every worker more productive, that’s the ROI, and that’s not actually true. The atomic unit of the enterprise is the workflow, you can make a single supply-chain analyst more productive and generate no incremental ROI, you have to make the entire supply-chain workflow measurably more productive to capture the value. There are some roles where the cardinality collapses — the knowledge worker is the workflow, like coding — so you have perfect product-market fit for those things, and you can chase unbounded productivity. But that’s a very small part of everything we do in the real economy. Again, I don’t understand the audience here. In a market, shouldn’t the model makers just do their job, which is making better models, and then there are lots of opportunities for companies to take those models and diffuse them? Isn’t that what Palantir is trying to do? Who is the target of this exhortation? SS: To those wondering, “Okay, we’re creating this, we’re spending trillions of dollars, where’s the return, how’s it working, why is the best example of excellence an AI assistant, which is nice and convenient but that’s not going to inflect our GDP?”. You think the big problems that I’m interested in solving — I think one of the fundamental issues, if you zoom out a lot beyond globalization. In 1971, the relationship between GDP growth and wage growth broke. Before then, every time the economy grew, wages grew. And now, adjusted for inflation, wages have been flat since. That’s a profound problem, a lot of our current challenges are downstream of that reality. There are many definitions of a thriving middle class, but the one that resonates with me is the most qualitative which is you believe your children’s future will be better than yours. If you go a couple generations without being able to cash that check, you get massive societal instability. Why does this matter to you? Why does it matter to my audience? Is this your mission here? Why do this interview? Is this concern about the overall public perception of the tech industry? Is AI going to lose public support? Is it about people opposing data centers? Where does the rubber meet the road, from theoretical discussion to actions you want to see happen in the real world? SS: Is AI just a dope technology, or is it something we can actually use to make our society better? I’m trying to solve for making the world better, our society better, through backpropagation . What would have to be true of AI for these things to be possible? As cool as AlphaFold is, there are zero drugs on the market because of AlphaFold. It’s dope — super dope — but nothing is better because of it, it would be a travesty to turn our back on AI in such a way that all it can be is something really cool. Number two of the Shyam interview outline, why are ambitious customers winning? Why do some units and institutions put AI to consequential use while others struggle to move beyond experiments? How much of the difference is access to technology, and how much is leadership, urgency, authority, and an imperative to win? SS: This is one of the empirical things I’ve observed. It wasn’t obvious to me it would be the case, but if I segment the customers I’ve worked with — the ones who want to use AI to reduce headcount, to take costs out of the business — almost all of them have failed at even achieving those goals. If you compare it to the customers succeeding wildly, have real ambition, “I want to use AI to dominate my market”, “I want to use AI to put my competitors out of business”, “I want to use AI to do work that was always beyond the efficient frontier — things I always wanted to do but could never afford to do”. I think the conclusion I’ve come to is that there’s insufficient stimulus to motivate you to do the hard things when you have a myopic goal like getting rid of humans. There might be other factors too — human intelligence and machine intelligence aren’t exactly the same thing. How so? How are they different? SS: They don’t do exactly the same things in the context of the workflow. If you spend your time trying to delete the human, by definition the AI would have to be able to do everything the human does — even if it only does 98%, you’ll fail at the specific goal. Another way to reframe it, “I have this miracle intelligence now, how do I build the workflow around what it can do, rather than shoehorning it into the error that always exists between the human and the AI?” — the AI maximalist would say, “I’m just going to keep training a better model”. But I’d say, what percentage of the problems we have today can we solve without going beyond the current frontier? I think we’ve left a lot of problems unsolved because we’re not thinking about diffusion enough, and wer’e not actually trying to maximize the AI for what it can do, we’re instead trying to judge AI by the standard of, “How do I eliminate humans?”, which may not even be that valuable, judging by the people who are succeeding with it. Does this lesson of the importance and power that comes from focusing on the top line versus the bottom line — was that a lesson you learned in the age of AI, or something you learned working with companies before current LLMs came along? SS: It’s very pronounced in the age of AI. Before then, you’d say it’s more exciting and you succeed more often going after the top line, but you’ll still succeed going after OpEx reduction. With AI, the outcomes are much more binary. It’s going to lead to a category of winners and a category of people who view this as something that’s not quite working. Number three, who carries the burden of systems engineering? A capable model is not a working system — despite lab blindness, the idea that there’s anything other than the model. Who owns the integration of probabilistic models with deterministic software, human judgment, permissions, and real-world consequences? Explain that. SS: Just this weekend there was another article about how LLMs are able to communicate using timing differences in VMs for CPU cache. Again, it’s a complete absence of a systems-engineering approach to the problem. It’s saying, “Can we test and find things that can go wrong?”, rather than viewing it the other way around, “What would we have to build around this so we could actually use it to capture the economic value that exists?”. There’s a ceding of human agency and responsibility around solving these as systems-engineering problems. The tell is you look at these historical incidents — in a real company, you’d expect someone to publish a “ Five Whys “, we’ve done a retrospective, I’m documenting not just what happened but everything I will do and build and fix and change so it can never happen again. Think about how [Hyman] Rickover built the nuclear Navy. These two positions are interesting to juxtapose, Admiral Rickover building the U.S. nuclear Navy, and the Soviet nuclear fleet. The Soviets had to rest their submariners every six months in Sochi so they could regenerate their white blood cells, because they treated safety from radiation as an unsolvable problem, “It is what it is”, and Rickover designed our submarines to a standard of safety a hundred times more than what scientists thought was even needed. Very clearly there you have a human who absorbed the burden of systems engineering, and a philosophy of continuous improvement. I think companies are going to have to absorb that, and the people who provide AI are going to have to absorb it too. All of these problems are solvable. We could wring our hands and talk about how dangerous this new thing is — we could have done that with the automobile. Four million Americans have died since 1899 with the invention of the automobile, but we just kept going, we have seatbelts, airbags, anti-lock brakes, there’s a systems engineering approach to reducing risk, dare I even talk about the airplane. You don’t see that coming out in the discourse — other than, say, Power Shelf from Nvidia , who has actually released toolchains meant to provide more safety? There should be a proliferation of that right now. Do you think that falls on the model makers to do that? Or does it fall on companies like Palantir or Microsoft? Or to what extent is it in the interest of Palantir and Microsoft to frame it that way because suddenly the value of what you’re wrapping around the models becomes much more apparent? SS: Well, certainly that’s the business we’ve ended up building, so you can say there’s some self-interest there. But I’m kind of gaslit by how little focus there is in the labs on this. It comes down to model blindness, there’s an aesthetic where if you need anything other than the model, that’s somehow a failure of the maximalist version of AI, and we should just be building more there. If you were more focused on the practical realization of benefits for your customers, you probably wouldn’t think about it that way. To what degree are the model companies really selling progress, not product? Isn’t that good for you? SS: It’s great for me! Well, here’s a question. You talk about the need to focus on the top line rather than the bottom line, and the need to build things around what these models are capable of as opposed to retrofit them into a human replacement paradigm. How does that apply to Palantir broadly? You built this business, this concept of the ontology , all these disparate data sources, and you bring them in — and when we talked in 2023, it was a very compelling story about how this makes sense in the context of AI, with AI almost being a user interface across this data. At what point are you approaching AI and it seems awfully convenient that, “We’ve had it right all along and it fits right in” — especially in conjunction with a message that “You need to rethink what you do around AI”. To what extent has Palantir rethought what you do for customers because of AI? SS: Enormously. On the big macro things, we were right. The enduring value of ontology — I was just with another big customer who was reasserting how ontology is what prevents the kind of AI slop from accumulating in their enterprise, and provides a layer they can build their supply chain on. But it would be crazy not to think AI has gotten us to rethink lots of small things, including how the work gets done. The majority of work in the platform now happens through what we call AI FDE — using the models themselves to accelerate the programming of what’s happening in the platform. Also thinking about it as part of a toolchain. If you zoom out, we view the enterprise as a state machine. Before AI, every one of those state transitions, some human was doing — the state machine is very complicated, and you’re interacting with lots of legacy systems. So the point of the ontology was never just a semantic representation of your data in your enterprise, it was also to model the verbs , the actions you could take. It made the enterprise programmable — a harness for humans to the enterprise. The opportunity, and what we had to invest in, was turning that harness into something that models can use. And thinking about it idempotently, enterprises aren’t just going to rip out the humans and suddenly have AIs running the entirety of it. Some workflows will start with humans, move to AIs, come back and forth. So you need a state machine that’s invariant to how did this work start — mammal or AI — and how is this work going to end, and how I govern the transitions through that state machine accordingly and the actions that can be taken? How would you have changed your approach to the ontology and what you’ve built had you known back in 2006 that LLMs would be the ultimate interface to what you’re doing? SS: Since the core ontology is kind of enduring, I’m not sure we would have changed that. It has changed how you get value out of the ontology — the applications you build on it, how they relate to each other, and it’s changed how you get data and model systems into the ontology. The counterintuitive thing, maybe a-year-and-a-half ago, I mandated that every new employee learn to use the platform through AI FDE in their first week, “This is the future, let’s go maximalist on it”, and it actually had a profound retarding effect over time, which was that there were lots of young employees who only knew the platform through this abstraction layer and, as a consequence, didn’t understand the underlying building blocks that the platform used and was composed of. Perhaps the most important question, even as someone who is an AI maximalist, is knowing when not to use AI. It’s been very interesting, when you ask the AI to design ontologies for enterprises, Going back to the alpha-beta frame — when you ask the AI to design ontologies for an enterprise, it does a mid job. Going back to the alpha/beta frame, it’s all beta, it models you like it would model anyone else, smoothing over everything that makes you special or different. A reversion to the mean. Number five — we’ll come back to four — has enterprise software made companies more alike, and will AI make that worse? This is an argument you’ve put forward on multiple occasions, the idea of conventional enterprise software as beta, not alpha — software required you to sell to as many folks as possible to get the return on your investment, and if you’re selling to as many folks as possible, it’s defining workflows, so everyone ends up with the same workflow. Isn’t that fine? Aren’t most workflows non-essential to the top line — software is inherently a bottom-line endeavor, and in that case it’s fine? SS: In the old world, that’s the only thing you could use software for. Go back to the earnings calls during COVID — what were CEOs talking about as the technology successes? Zoom, Teams. Meanwhile, many of these companies who I’ve known had just spent five billion dollars on an ERP implementation that fell over like a paper tiger two weeks into COVID . So the question is, what parts of the software-industrial complex are actually working? I don’t besmirch them — essentially they’re solving for, “Can I sell you the software?”, which is not the same thing as, “Did the software work?”, “Does the software actually matter?”. You can juxtapose that to our entire approach to software has been solving for, “What software must exist to solve the problems I care about?”, it’s building software through backpropagation. That both lets you continuously up-level the ambition of the problem you’re trying to solve, and allows you to think through the abstractions you want to codify in software versus the parts you leave as what is going to be unique to the customer. That’s allowed us to pierce the trade-off between alpha and beta, “ How do I build software that makes each customer more different, not more similar?”, and therefore is a strategic weapon they can use in the market. So should customers be using Palantir for driving top-line revenue and still hiring a bunch of SaaS companies for everything else? SS: Empirically, that’s the case. No customer I work with is a greenfield where we’re the only thing that exists. If you’re happy with the beta you’re getting but you’re beating your head against the wall trying to figure out why the beta doesn’t let you capture the opportunities in front of you — are you fighting your software stack? Are you fighting decisions made in the past, borne out of an aesthetic desire for something easily governable and maintainable by your IT organization? IT will always yoke toward beta — decisions that are mimetic, closer to best practice, what everyone else uses. The people most interested in alpha are in the line of business — the person who feels frustrated that their strategy is hard to implement, versus “I have a precise steering wheel to do this”. So do models deliver beta or alpha? SS: Depends how you use them. If you just think about it at the limit, we’re just going to keep betting that these models will get better and I’m going to rent my intelligence from a lab — how are they getting better? They’re somehow compounding all the world’s knowledge into their system, which you’re maybe tacitly helping them do that, maybe you’re not, but that’s the trajectory you’re on. There should be an existential question about, “What is the theory of my firm?”, “What do I do if all the intelligence is compounding into weights I don’t control?”. A big part of what we’ve been pushing is this idea of Sovereign AI , that our customers have really taken to. It’s not a binary — you don’t need to leave the lab ecosystem. But for what workflows is actually the most valuable asset your post-training pipeline — it’s not about a specific open-weight model, it’s not about a specific post trained open-weight model, it’s actually that you have sovereignty over a post-training pipeline filled with the operational exhaust of your enterprise. The operational decisions that you’re making in the context of the operating system that is our software is generating data that allows you, every month, to train a new checkpoint and evaluate it against everything that came before, in a way where you know that model is yours that you’re going to be relying on. The stakes of doing that in something like government were obvious, your competitive advantage is everything, you wouldn’t rent the brain from someone that your competitors could rent the brain from. But it’s equally true in commercial world where frontier capabilities seem to be a function of unique data that pushes you past the frontier. If you talk to the Mercor’s or the data companies of the world, they’ll tell you the way they sell to a frontier lab is they generate a dataset that they post-train an open-weight model on to be provably past the existing frontier of the model the lab is selling. So if that’s how this game is happening, every company has this exhaust as a unique asset, why should they sit around and wait? And then as a consequence have a bigger question of, “If everyone has access to this, what is my alpha as a firm? Is everything reduced to beta?”. That makes sense — you give everything to an Anthropic or to an OpenAI, that learning is going to benefit everyone, which is going to transfer your differentiated knowledge into beta instead of being alpha. But why is the correct stopping point having a partner like Palantir? Why shouldn’t they just do all of this themselves? Why is it that you are a trusted partner in a way the models are not? SS: You can think of it as a model factory. What systems are you using today to make operational decisions? What system is generating the exhaust that has your decision-making processes and all the constraints that is effectively its own RL environment that you can go off post-train these models in? And as you post-train those models, how do you evaluate those models against historical decisions you’ve made, how are you going to manage deploying them to production? Having the entire vertically integrated chain where it just works and you’re able to leverage — a lot of what we’re investing in now is auto-research, how do I, without the customer investing in a big post-training team, use the decisions they’re making in supply chain or in production planning, use that exhaust to automatically start training models behind the scenes and surface to them, “Here’s a model that’s better than what you’re doing”? How do I use every point where you’re using an LLM and use that as a trace that I’m collecting that feeds back on itself, so that you have this continuous improvement? That’s the compounding loop, I think every customer wants to be in it. How do customers think about timing? It might be the case that in the fullness of time you’re just getting a median answer from a probability distribution, it is all beta, but if you’re first to beta, you can still make a lot of money. Is that the sentiment that you’re running like, “If we incorporate AI first, sure it’s all mid, but timing matters too?” SS: To make sure I understand, the mid point is coming from the frontier labs? Or weaker post-trained models? Let’s say you’re in a company that is considering it’s competitive set, who they’re competing with and they’re like, “Okay, Palantir, this is a good pitch, I think in the fullness of time you’re right, we should not be giving our knowledge to the frontier labs, we should be building our own hill-climbing capability, keeping it all internal, that’s what differentiates us. However, that’s going to take a while to get that set up. I could lean in to using the models right now, my competitors are not, if I’m first, I’ll have a short term advantage, I’m motivated by the short term, and that’s going to be enough”. Starlink on airlines, everybody is going to get Starlink eventually, but United is going to get a lot of benefit from doing it first. SS: I would tell customers they should win today, the primacy of winning does matter . But the part I push back on is, “How long does this stuff take?”, I’ve actually been surprised — it takes a week, especially if you have the operational exhaust where people are already using you to make the operational decisions, to develop a new post-trained model you can put into production. So that’s your alternative. Empirically, where we’ve done this, you get better-than-frontier performance at task-specific things. Say my business is selling electronic components, when people submit a request to purchase components, they submit an electrical-engineering diagram — pretty technical, definitely out of distribution for the frontier models, it requires a vision-language model. I have all my historicals, what the input is of the spec, and what the BOM [bill of materials] was that I quoted back out that a human built over the last 10 or 20 years. It takes a week to go train that model, and now you’re able to respond to your customers with a price faster than your competitors can, because it’s going to tak them weeks to work their way through the same specs. Presented by When I write about tech companies, I am often writing about businesses that depend on Stripe , because Stripe provides the foundation for Internet-native companies — including Stratechery. Consider Substack: building Stratechery was complicated and involved gluing together disparate pieces; Substack made it easy by delivering an integrated offering. One piece that was the same, however, was Stripe: Substack creators are also Stripe customers, as are creators on many other platforms. Zoom out and the pattern repeats. Fintechs can build entire business models around card programs without ever having to build the issuing and processing infrastructure underneath it. AI companies can charge for what customers actually use, which is particularly important when usage also drives costs, without having to build the systems to track and bill for it themselves. No business succeeds because of Stripe. Stripe doesn’t answer the hard questions about what to build or whether anyone will actually use it. But it does make more things practical to try. Number seven, could successful AI adoption be bad for frontier-model economics? If a workflow succeeds on a frontier model and then migrates to a smaller model the customer controls, or to ordinary software, who captures the value? SS: I think the customers will capture the value. I don’t think it has to be bad, but this is the fundamental question going back to something I said earlier which is, to what degree are the frontier labs selling progress that the next model will always be bigger and more powerful and more interesting than the prior model, versus to what degree are they selling a product? The more they’re selling progress, the more customers are rationally going to take non-frontier use cases off their workflows. Is it inevitable that the labs have to come for all software companies, because of the commoditizing nature of their models ? SS: I don’t know if it’s inevitable, it’s an open question, there’s still a question of how much value do you realize from incrementally more powerful intelligence? Right now we’re in a peak phase of not being able to distinguish between things that are dope, every new model release, I love, it’s really cool, but I’m also not seeing a commensurate amount of translation of new economic value unleashed that I couldn’t have previously gotten before. Why won’t frontier labs just own the entire inference layer from top to bottom, from smart models all the way down to dumb models? If it’s ultimately a commodity market, won’t they have a superior cost structure given they have the best AI to build the best inference anyway? SS: If it’s just about the cost of inference, probably. The question is whether you’ll have more and more things that are out of distribution, and that you get better performance from models you own, and to what degree companies will be comfortable on use cases that do represent alpha? So is this the key thing? Is your reason to be out here to exhort companies to stop giving their data to the model makers because, all data being equal, they will win? SS: I definitely think that’s true. Partly it’s to make sure companies know there’s an alternative for the use cases that aren’t solved today and to the extent you’re wondering where exactly is the ROI going to come from, why is it that the only uptake I’m really seeing is coding? I’ve had these two out-of-body experiences, probably bad form to talk about, with each of the frontier labs, where different people were saying, “You should really see the work that we, the lab, have done at this company”, but the reality is it was work I did. Yes, I’m spinning the token meter, but the translation of the model to economic value for that company required someone who actually had a sharp opinion on diffusion and the realization of that value. Is DeepSeek the most effective Chinese company there is? Particularly given their MIT license, and they just open sourced their inference stack . Do you use DeepSeek, how do you refrain from using it? SS: We don’t use it ourselves, but we let our customers in the commercial world have choice in the models they use. While I encourage them to use the Western stack for lots of reasons, and there’s a burgeoning Western ecosystem that’s definitely good enough, there’s a degree to which people have over-indexed on benchmarks and the bench-maxxing that’s happened. When Nemotron 3 was released — just vanilla, non-post-trained Nemotron 3 — we brought it into our own stack that we use to manage our internal operations, and there were five use cases where it was better than frontier, which should not be possible if benchmarks were a true, faithful representation of intelligence across its full jaggedness of it, should not be possible. So I lost confidence in benchmarks, and started this movement internally that it’s not about bench-maxxing, it’s about bench-making. The cardinality of benchmarks should be at least one-to-one with the number of customers in the world — what is your benchmark, what do you care about, how do you evaluate models against those things you care about? That’s a very empowering frame, getting out of, “What’s the latest, coolest model?” versus, “What actually matters for me, how much improvement is there, how do I trade off latency, cost, performance against all these dimensions?”. Why don’t you make your own model? SS: If I felt we had the need to make one, I would. Mostly since we exist in the context of our customers, I help them post-train models that solve their problems. I’ve toyed with and think there might be opportunities to create models more intrinsic to our product, that are distributed to all our customers that solve very specific problems, or a starter checkpoint they can start off of, that I think gives them a running head start against some of the problems they care about in domains with unique phenomenology in particular, things way out of distribution. I could see your overall argument, we’ve talked before, and we both love looking at industries in terms of the value chain and where it actually matters. I think your overall argument, and I’m trying to play Devil’s advocate here, I absolutely see the value in the data and owning the ontology, that and chips being the inputs all resonates with me. At the same time, there’s an aspect of controlling your own destiny, the essential inputs, and you opened this discussion talking about the problem with globalization — the idea that on the margins it made sense, but in totality you lose the plot. It’s interesting to the way you’re concerned about the models whether that be from an ideological perspective, from a them taking your exhaust perspective, from a them backwards-integrating into what you do perspective, why aren’t you investing to have an alternative, or contributing heavily to those who are? I’m sure Mark Zuckerberg would take some money, I’m sure Jensen [Huang] would take some money. Is there an aspect of putting your wallet where your mouth is? SS: The business case would be, I’m not sure the value accrued to the models. Certainly the residual compounding value is what we provide our customers. But then what’s there to worry about? It’s not going to accrue to the models anyway, isn’t the market going to take care of all these concerns? SS: I’m not worried — I think it will. Partly my exhortation is around, if we’re in a rush to realize the value of this genie of AI, are we actually putting all our emphasis and chips in the right place? We should be careful not to think of the Chinese as being ten feet tall but if you juxtaposed it with the profound focus the Chinese have on AI is about productive forces — how do we actually get economic value out of this? You can say that even the commoditization — one there’s an aspect of economic warfare through distillization attacks, we’re going to release an open-weight model and destroy a huge amount of U.S. market cap but the other part is, it serves as a stimulus and subsidy for the rest of their ecosystem, where they want to get better or are really good, from power to chips and I think it won’t be that long before the Chinese models are not open-weight. Why is that? SS: Because as they start to feel they’re at the frontier, or attempting to get beyond it, they’re going to try to monetize it and close it down. Do you think they’re at the frontier? SS: No, they’re behind it right now. Can you be a Chinese model maker and not be totally dependent on distillation? Doesn’t that leave you completely uncompetitive with your peer set? SS: Yeah, it does. Which maybe is the other aspect of it is how the CCP views the models as a threat to their own security and stability. There feels like there’s an aspect on all sides of, “Let’s just trust the market and our intuition”. From my perspective, I wrote Who’s Afraid of Chinese Models? this summer, which got at the fact that everyone is stuck on API prices, which of course include a profit margin. At the end of the day, I want to know the actual cost to serve. My assumption, I think a reasonable one, is that the frontier labs can serve very efficiently, in part because they have that capability, serving a very large Chinese model costs real money, it’s not like you can serve it for free. By the same token, this assumption that because they accelerated to six months behind, or three months behind, whatever you want to say, that it’s inevitable that acceleration will surpass — when the distillation explanation both covers exactly what has happened and continues to cover what has happened. When you think about the overall economic forces of the Chinese labs, they have to depend on distillation, if you develop a dependency on distillation, you won’t develop the capabilities to be on the frontier. So why can’t we all just relax? SS: Because winning AI is more than having the most powerful model. Maybe this is the fish-in-water thing I didn’t articulate, why do I have all this angst about how quickly we’re diffusing? Because that’s the whole race, the whole race is who realizes the most value and is in the best compounding loop around the things that matter, as opposed to the dope fact that we have a cooler and cooler, more powerful model. Got it, I finally got you to answer my first question, I just had to go around for a while. Are the labs copying forward-deployed engineering, or just the job title? Palantir is very famous for this — you specifically, the most famous for this — what is a forward-deployed engineer as you developed it, and does that apply to what the labs are doing in terms of creating these organizations that go into companies and help them adopt AI? SS: At this point it’s turned into the most popular job title, it goes beyond the labs, really. What a turn of fortune, huh? SS: For 17 years I was maligned as the person ruining Palantir because I created this horrible concept of the FDE, and then it became the coolest thing ever. But the challenge is, before you say, “What is a forward-deployed engineer?”, you have to say, “What is forward-deployed engineering?” — and that actually it’s a product strategy, not a role. There’s a role, yes, but it’s way downstream of the thing that makes it happen. The reason I think it’ll be very hard to copy is there are probably five people in the history of Palantir who know enough about the strategy to actually recreate it anywhere else. You can’t just go hire someone who was an FDE and do this. Just even as a starting point, forward-deployed engineers are not part of your go-to-market organization, they are part of your product organization. Right, they’re deciding what gets built by seeing exactly the problem the customer is having, communicating directly or being part of the product team and that influences what is rolled out next. SS: It goes through multiple layers, you’ll have to indulge me as I try unpack it. The first part of it is if your product exists on some Gaussian distribution of features, “I’ll average the signal across my thousands of customers and decide what to build next”, forward-deployed engineering is not the right strategy for you. If, on the other hand, your customer set exists on some power law, where for a set of problems you’re solving, there’s a small number of customers with the hardest version of the problem and if you can solve it, you get the entirety of the market, because everyone else will have this problem, they just don’t know it yet, some are living in the future — then you have to go from specific to general. The journey from specific to general is very hard is very hard, engineers hate the notion of wasted work. But you get from one to two to three, and you’ve earned enough surface area to understand the general form of the solution that’s going to work for the fleet. It’s a heterodox way of thinking about strategy, and it requires thinking about the shape of the engineers who do this work differently. Engineers exist on a broad continuum — this is a caricature, intentionally so, to be illustrative — on one end you have the hacker, the MacGyver, the person whose dopamine hit comes from solving this problem in the world. They’re the best duct-tape users in the world. SS: Absolutely! Never mind, it’ll only work once or twice, but, “This thing is true because of me”. They have a knack for figuring out which problems are worth solving, which is hard, let’s call it the first 80% of the problem. There’s a different skill, call it the artist, the other end of the dialectic, the engineer whose dopamine in life hit comes from how beautifully built the thing is. Never mind it solves no problem anyone cares about, “Don’t you see how elegant it is?”, this dialectical tension between two extremes, and how you get to the second 80% of the work. Each personality undervalues the contribution of the other, so it’s held together as a dangerously unstable equilibrium that has to be managed. I think we’re up to 96% if I’m doing the math right, so what’s the last 4%? SS: No, 80% + 80%, the whole point is the math doesn’t math, because everyone thinks they’re doing the majority of the work. So the answer is, someone sees a problem, doesn’t just understand the problem and what to fix, but how to fix it systematically so that it’ll be a sustainable fix not just for that customer, not just that time, but going forward, for all customers. SS: Yeah, and earning the opinion on where the abstraction should sit in the stack so it can actually generalize because building an end-to-end solution is overfit to one customer. So what are the parts you’re actually going to generalize, and what parts do you let be customer-specific? It’s engaging with an inherently messy process. I think a lot of folks are jamming a sparkling form of sales engineering, they take the FDE role and say, “We have really smart engineers who can go work with customers”, but there’s no way of metabolizing the pain the FDEs are experiencing into product, if you don’t have that, you don’t have FDE — it’s just sparkling sales engineering. Two questions to follow on this. When you talked about the Gaussian distribution of customer problems that lends itself to, “Build one thing that broadly solves a problem for everyone”, in many respects that’s consumer software. The best consumer software companies and consumer software leaders do not rely on focus groups, because you can’t get a focus group on eight billion people — you have to have your own sense and drive the software in a particular direction and believe in your vision. There used to be a phrase, you’re old enough to remember this, the consumerization of IT . Was the consumerization of IT, in your view of the world, the application of this “one solution for everyone” mindset, when engineering IT back in the day was line-of-business apps, it was people actually building software for their organization? In the SaaS-ification of everything, the spreading out of R&D across as many customers as possible, that is the issue that you’re pushing back against, and that’s actually what happened with the consumerization of IT. It wasn’t just about user interfaces, it was about one-size-fits-all. SS: I’d agree with that. But I think the correct point is an unstable equilibrium between the two. There’s a constant need for the masses to revolt against the high priests who have their perfect conceived notion of how things are supposed to work but they don’t work. Part of the problem is there was an economic rationale for this in that developing software was exceptionally expensive, so you needed to spread it across everyone, and because you had leverage over enough customers, you could do an 80% job that was better than a dedicated individual. I think the argument for AI is actually no, you can do small-scale engineering projects that solve 98% of the problem instead of 80%, and get back to a “many flowers blooming” approach much more akin to the ’70s and ’80s in terms of enterprise IT. SS: I think that’s right. And again, you’ll find the pendulum keeps swinging. There’s a proliferation where people go all-in on Codex or Claude Code in the enterprise, a proliferation of apps that don’t speak to each other — almost like attritional software. You built it to solve a problem, you either kill it or you accumulate debt over time. So what constrains that entropy? That’s what we think ontology does. The ontology provides that constraint point you can govern, so these things can interact, you can have a lot of freedom to vibe-code an app, and the cost of it operating in your enterprise is low, the cost of throwing it away is low. So you can empower your builders without giving up your ability to have a cohesive operating system for the enterprise. Here’s a defense, or a question, about the AI labs putting forward their versions of FDEs — or was it “sparkling sales representatives”? Isn’t the actual FDE the model itself? It ties into your overall concern, the way for them to solve your problems is to actually absorb all of your data. So yes, the people are not the FDEs, the AI is the FDE, and that’s the actual FDE you’re worried about? SS: I’m actually not worried about that now, this goes back to my experience of, “Let’s just have AI FDE be the entry point for everyone in our platform”, it’s all a mean reversion. There’s a lot of value in getting the problem decomp right, and if you get only an okay problem decomp, you leave a lot of the alpha on the table. That’s probably the part I’m least worried about going away. What are you most worried about? What has changed between 2023, when we last talked, and 2026, today? Are the models more capable than you expected? Less capable? How do you view the world differently now than you did before? SS: The models are more capable. I don’t know if I had an expectation — it was so hard, in a world before AI, to truly have an expectation, but it’s amazing, I’m delighted every time I see a new model release and get to play with it and get to see what’s possible. What I’m more convicted of is that there’s not a one-for-one overlap between human intelligence and machine intelligence, and the more we’re willing to build things that make sense for machine intelligence alongside the humans, the more value there will be. I see this in subtle ways. This might seem orthogonal, but it’s very relevant — I have this parts manufacturer for the submarine industrial base, and it used to take them two weeks to do their production plan during which the line was down. With AI, it took less than ten minutes, and now they have all this excess demand they didn’t have people to go do so they hired a third shift. There’s an army of blue-collar laborers who have their job because of AI, not because they use it, not because they think about it, they’ll never tie that back together, it’s the opportunities in our economy I don’t think people are going after hard enough. Just one question, I just mentioned the Anthropic thing — a lot of this feels, at the end of the day, feels like a lot of this conversation is ultimately about Anthropic. You were very close partners with Anthropic — the whole using Anthropic in the U.S. national-security apparatus thing was through Palantir. Do you regret that? When the whole showdown happened with the Department of War , were you stuck in the middle, or decisively on one side or the other? SS: Ben with the tough questions. Yes, we were stuck in the middle. I don’t regret it. We still work with Anthropic, we work with both frontier labs. This speaks to AI sovereignty right on the head which are who gets to decide how these things are used? I do think there’s a question of epistemic humility over these issues, which I’ll get to in a second. The customers have to decide if they’re going to be okay with the terms of use of a model or not, and what will the market accept ultimately. I’m not sure it’s going to be sustainable for model companies to have long, nuanced arguments over, “How will you use this thing?”, it’ll be much more binary, do you want to work with this institution or not? You have to trust your customers, to certain degrees around these things. It was probably not handled excellently by many parties involved, but it did surface an important issue here, when the intelligence is coming from one place, what are the conditions do you need to know you can depend on it in a mission-critical environment — whether that’s the factory floor or a government organization? Did that change what models you’ll work with — or is that still up to the customer? SS: I want that to be up to the customer. What it did change is we spent a huge amount of energy and effort on being able to safely swap models out , and being able to exercise that sovereignty. You need that just as much to go between two frontier labs as between a frontier lab and an open-weight, post-trained model that you’ve created. Creating that scaffolding gives customers assurance, if you thought it was a two-year journey to do one of these swaps, the decisions you’re making are effectively a one-way door. If it takes two minutes to do safely against an eval set for a specific set of workflows, you don’t have to be hung up about this in the same way. I think I’ve created a lot of trade space for my customers as a consequence. What was a bigger effect in developing this capability — that episode with the Department of War/Defense, whatever you want to call it — or was it that whole Fable “ No more zero data retention , I promise it’s for safety” angle? Was that the one that actually woke people up? SS: The second one woke up the broader commercial world, because until then it seemed like some narrow government dispute, “What do I care about that?”. The first one allowed us to build the product that we needed to enable that so when the second one happened, we were able to give our customers choice, to make decisions inside the OODA loop of “Holy shit, I just woke up to this reality, what do I do?”. Did that apply to you? “Holy shit, we just woke up to this reality — what did we do?” SS: Oh yeah, a hundred percent. So was that actually the best thing that happened to you? SS: I’m not sure if it was the best thing that happened to us, but we felt like we had seen where the world was going, and the sorts of trade space we would need. There’s a clarity that’s come, I think, from Alex [Karp] and from Satya [Nadella] in particular , who have basically been saying the same thing. Maybe I should ask you, is Microsoft actually your competitor in the long run? Both of you are talking about this concept of being harness-makers — the ontology, Microsoft Graph, whatever you want to call it, having the data in a controllable way, building your own hill-climbing capability, so you’re absorbing your own learnings rather than giving them away to a model. But it felt like I wasn’t getting that message clearly from either of you until about six months ago, and it’s like no one was really paying attention to this until then. SS: I’ll give you the inside baseball. Before Alex’s CNBC interview , it was the end of the quarter, he’d probably visited sixty customers, and every customer in private was telling him this, but none wanted to say it in public. It was very clear that there was this underlying frustration and belief that no one was willing to put their finger on. That’s part of why it had so much resonance, and why he delivered it with so much conviction and certainly, if you look at what we’re saying, we see the world in very similar ways to Microsoft. Number four, I told you we skipped this one, has effective altruism turned AI safety into a secular religion? SS: Clearly I think so. One of the things I’m worried about most, this goes back to the systems-engineering critique is that you have this weird rhetoric — the argument is essentially, “AI is going to create utopia, but of course there’s some chance, between 0 and 20 percent, of ending humanity”, so dystopia. These seem contradictory— They’re not contradictory — they’re two sides of the same coin. It’s a rejection of any middle pragmatic path towards absolutes. It’s extremism, just manifested in two different ways. SS: I’m preaching to the preacher here. It rejects all human agency in what’s happening right in front of us and therefore the obligation toward systems engineering. It influences how you even think about testing these things, what you think would have to be built around these things. I don’t begrudge anyone their articles of faith, but when your articles of faith are represented to the world as technical critiques, it’s very dangerous and risky. People think, “Who could know more than this person who works on this thing?”, and that would be true if it were a technical critique, instead of a religious perspective that comes out of a unique worldview. It’s a combination of a millenarian cult with a messianic complex. It’s, “This thing is going to end humanity, except if I do it, we have some chance of working out”.** You are preaching to the choir — I’ve written about this. You have to reject the frame completely . You don’t argue the relative chances, the expected value of utopia versus dystopia, you question why those are the only two options that might come out of this. But here’s the question: for many years people would say Palantir is maybe akin to a religious organization. Today defense tech is hot, but when Palantir came along it was not hot, you went there because you were a true believer. We may have talked about this last interview, we’ve certainly talked about it at some point, and the power that granted you as an organization to push through all these problems, to figure out all these issues, because you believed you were doing something important and meaningful. As you look back, how important was that belief system — the shared belief system — to Palantir’s success? SS: I think it was crucial. Part of the reason I can see this so clearly is that cult recognizes cult. We were a very different sort of cult, in the service of the West, in the service of democracy, a belief in state capacity, that the legitimacy of our institutions depends on the people who been elected to run them having the tools to run them in a way that serves the will of the people or their shareholders. That national security and prosperity are two sides of the same coin, national security is not an end unto itself, it’s a means to prosperity and human flourishing. You’ve got to get these things right, particularly in a post-terrorist world where every one of these events was a one-way ratchet toward less freedom and less security somehow. So yes, it was crucial to getting us to work, that’s why cult recognizes cult, although they’re very different cults. Is your critique that this is a religious movement, or that the religion is wrong? SS: That the religion is wrong. So this is a religious war. There’s a War on Terror callback for the people who know. SS: It is a religious war, but I’m not interested in fighting the religion head-to-head, I’m interested in making sure that people listening to the doomsaying have an understanding whether they’re hearing a technical critique — the presumption everyone has — or a religious perspective on a future world that may exist, and then let them decide. How is it the case, though? It’s really striking, I’ve been picking at you and teasing at you that all your arguments, which I believe are heartfelt, do happen to perfectly align with Palantir’s business model. It is in your interest for companies to understand and believe that the data layer is what matters, they need to control that, that the models need to work in service to them, not throw your exhaust to the model, you’ll get the median distribution anyway, etc. I think that’s all valid, and it also perfectly aligns with what’s good for you. By the same token, the remarkable thing about Anthropic — which just filed their S-1 , I haven’t had a chance to read it yet — but all their fervor, doomerism, and the things they put forward that you’re critiquing for not being technical, happens to perfectly align with their business model . It aligns with what they’re selling, and, interestingly enough, often with their customers’ interests — their customers don’t view Anthropic as a competitor, “Better they’re trying to create God than trying to compete with my SaaS application”. Is there a certain extent of game-recognize-game here? SS: The part I’d say is we’re all still founder-led institutions with a huge amount of agency over our business strategy and what we’re going to do. And the best way to run that is to align your strategy to things you actually believe in, all these things are all so hard. So I think the business is downstream of the fundamental beliefs you have. So at the end of the day, what decides who wins? Is it who believes harder, who works harder, or will fundamental market forces come to bear? You’re making a lot of arguments that companies that don’t pursue the median, that preserve their businesses, are going to succeed in this world, and in which case, isn’t everything going to be fine? SS: Unless we engender non-market forces that are reactionary. Like wrong safety regimes, a drive toward nationalization, we have data-center moratoriums that are engendered through a doomerism belief structure, we’re not able to rationalize and execute on delivering the benefits of AI to the average person in a way that lets them participate in the economic upside. All these forces, everything from a replay of the French Revolution to turning our back on the fundamental progress of capitalism are the greater risks. Shyam Sankar, thanks for the interview questions. SS: Thanks, Ben. Thanks for having me. This Daily Update Interview is also available as a podcast. To receive it in your podcast player, visit Stratechery . The Daily Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly. Thanks for being a supporter, and have a great day!

0 views
Stratechery 2 weeks ago

Meta Earnings, Muse and the Frontier, Musebook

Meta's earnings come in the context of Zuckerberg delivering on his promise; if investors believe in Muse they must support Meta's spending on compute.

0 views
Stratechery 2 weeks ago

Anthropic S-1, The Cost of Compute, Who Is Paying?

Anthropic's IPO is finally here; the core business is very profitable, but competition may eat all of the profits, even as investors are set up for endless dilution.

0 views
Stratechery 3 weeks ago

The End of the Internet

Listen to this post : It was the summer of 2003, just weeks before I was set to move to Taiwan, that I decided to create a blog to document my journey. I had, at that point, had a GeoCities site, a Blogspot site, a site I built in Microsoft FrontPage and hosted…somewhere, so this was my fourth swing at the blog thing (it was the seventh that ended up being Stratechery). I was intrigued by a just-released piece of software that promised a five-minute install; it was called WordPress. It did, I must confess, take me much longer than 5 minutes to install WordPress: it turned out I needed to learn how to create a MySQL database and configure PHP; once you had all of those component pieces in place, then yes, it only took 5 minutes to install! I was thinking back to those days last week when news broke that every version of PHP had a critical bug that allowed for remote code execution on the host machine; Stratechery today still runs on WordPress, but that WordPress is managed by Pressable , an Automattic company. That meant that my site was patched before the vulnerability was disclosed, and it would be crazy for me — or anyone else — to do it any other way. It’s kind of insane to think that twenty-some years ago I was doing everything by hand without having any idea what I was doing, and that it was totally fine. That’s the Internet, though: an insane invention from a time that is, sadly, gone forever. 2003 was, roughly speaking, right around the beginning of the blogosphere; yes, blogs existed — Dave Winer, often considered the father of blogs, started Scripting News in 1997, after early experiments with a personal site called DaveNet ; Justin Hall started Links.net in 1994; Jorn Barger coined the term “weblog” for Robot Wisdom , also in 1997, and the influential kottke.org started in 1998 — but it was the early 2000s when blogs suddenly seemed everywhere, particularly in the political space: That political blogs featured so prominently speaks, I think, to both the power and peril of the Internet’s first transformation. Up to that point the Internet was mostly technologists, comfortable discussing the intricacies of text editors and programming languages with other nerds on BBSs, early forums, and mailing lists; it was the politically minded, however, that realized that the freedom to start your own publication that could reach everyone was a radical development. The goal was not to make money; it was to make a difference. Originally these new publications were novel; what soon became clear, however, is that they were the tip of a wave of content that would dramatically expand the web. Newspapers had influence by being geographic monopolies; early bloggers gained influence by being the first to a new medium. Ultimately, however, the real winner of the web was Google: as the amount of content became abundant, Google offered not distribution — which was free — but rather discovery; indeed, it’s because distribution was free that discovery became the most important part of the value chain. Many of those early bloggers made a decent amount of money from Google through its AdSense advertising program; by the end of the decade, however, the ads became more intrusive even as the payouts shrank, both because the inventory was limitless, and also because the most attractive ad slots were on Google Search. And so we saw one of the first examples of how the Internet fails: because everything is open, everyone has an opportunity, which means infinite competition, which is terrible for business. For many years I rued the fact I wasn’t one of those early bloggers; a decade later, however, I realized that I was actually just in time for a much more sustainable business model: subscriptions. I wrote in 2015’s Blogging’s Bright Future : It turns out, though, that there are more business models than simply advertising, as Sullivan himself sought to show. Sullivan was, for the last two years, supported by reader subscriptions to the tune of nearly one million dollars a year, and while many were skeptical of Sullivan’s efforts when he started, his ability to raise money shouldn’t have been a surprise: his is a singular voice — to put it in economic terms, Sullivan’s writing had a low elasticity of substitution — so people were happy to pay for something they couldn’t get anywhere else. Right there was the exact loyalty that Klein fears has been lost… I am, of course, acutely aware that there is a tradeoff when it comes to the subscription business model: by making something scarce, and worth paying for, you are by definition limiting your number of readers. Stratechery, though, serves a niche, and niches are best served by making more from customers who really care than from milking pennies from everyone. In short, the answer to drowning in abundance is to purposely limit your audience — to go small. That is an economic model that works, and notably, there are a lot of bloggers on that list who are using that model today. Blogging is alive and well, it’s just constrained, and thus worth paying for. Google wasn’t the only Aggregator: a decade later there was a new means of discovery, which was social media. Now instead of searching for a site you could have a link shared directly to you, not just by email but on a site that served up interesting updates about whomever you chose to follow. And, for a period of time, followers could become friends. Stratechery’s fortunate timing was not just in terms of subscriptions becoming technically viable, but also the fact that Twitter was a community, not unlike the early Internet: genuinely interesting conversations happened in public, and you could reply-guy your way to broader distribution. Social media was also free, however, and even easier to get started than a blog ever was; that meant that the amount of content was greater by many orders of magnitude. This meant that algorithms started to matter — Facebook had one relatively early, but it took Twitter much longer — and also anyone with any sort of prominence had to start taking a much more defensive approach. Community, sadly, doesn’t scale; any platform that is available to everyone starts out special and spirals from there. Ultimately, public social media sites became personal entertainment sites, tuned for engagement, culminating in TikTok, a social media site with basically zero social component. However, that doesn’t mean social networking went away: in fact, I think that social networking is stronger than ever; the difference is that it isn’t public, but is rather in group chats. Group chats are familiar, they are high trust, they are, in many respects, the best of the Internet: even as technology atomizes us the group chat is the only place you can connect with people you care about, anytime and anywhere. In other words, the redemption of social media was — just like blogging — to become smaller. The scaling that mattered for most people was not an ever larger audience, but a proliferation of infinite finite groups. Social media 1.0 , even at its best, followed the 90-9-1 rule: one percent spoke, nine percent commented, and 90 percent lurked; the magic of group chats is not in giving everyone a voice, but a place where you feel safe to talk. The PHP vulnerabilities were just two of a host of security issues last week alone; there were also: This is, believe it or not, only a partial list, and I expect things to only get worse. All of software is built on an inherently creaky human-authored foundation; the human factor meant the primary gate in all of software development was “Did it work?”, not “Is it secure from every possible privilege escalation or race condition?” or the myriad of ways that software can act in ways not intended. The saving factor previously is that attackers were limited by human time and ingenuity as well; AI, however, has limitless time and the ability to substitute iteration for cleverness — and it’s getting pretty clever as well! It’s not a surprise that this wave of vulnerabilities is coming to light, and it’s going to get worse before it gets better — which I do, for the record, expect to happen; after all, the good guys have AI as well, and in the long run software will be better when it is (re)written by AI, and defended by agents continuously. At the same time, what this entails is that the public Internet, to an even greater extent than now, will be a battle zone at best (i.e. social media), and a wasteland at worst (everything else). Some of this is economic, as AI consumes everything — both content and attention — without giving any of the traffic back that Google did, but it is also technical: it was already a dodgy proposition to stand up your own site or app, which was the very promise of the Internet, given the hacking risks; in the age of AI it’s madness. It is sad. When I was a kid we called up BBS’s that were hosted in people’s homes; when I was in high school we downloaded music from public FTP addresses; when I was in college we set up our own websites. Anyone could set up a server, and that server was just as much a part of the Internet as any other server; that’s simply not viable any longer, even if you weren’t worried about making money. The Internet as we knew it is dead. There are, however, green shoots of possibility, and once again the answer is to go smaller. When anyone can make an app, the answer is not to try and make a quick buck in the App Store; in fact, the surge in slop will make the market worse for everyone. Rather, it’s to make something for yourself and a small group of trusted friends and colleagues. This is exactly what has happened at Stratechery: we have made a bunch of different apps, for everything from podcast production to task trackers to calendar management, none of which will ever be made available publicly. That doesn’t even include the apps made for oneself: I’ve created a bunch of different apps to manage my house and sports viewing: it turns out I had a lot of software-shaped holes in my life that suddenly I can fill without asking permission. Software has always been different from physical goods in that its zero marginal cost nature meant it could scale to far more people; at the same time, the sheer amount of up-front effort involved in creating it meant that the intended audience needed to be large enough to justify the investment. AI completely changes that: suddenly it’s possible to create software for one person, or one team; smaller target audiences aren’t just a necessity for security, they’re an entirely new and potentially much larger market for software creation at scale, by literally everyone. Or, to use the example above, the group chat era of software is here and it’s going to be so much better than what came before. Venkatesh Rao wrote an absolute must-read essay over the weekend called Cognitive Mercantilism . In brief, Rao makes the case that mercantilism rested on an assumption that economics was about scarcity and control; what brought about capitalism was demand. When people were freed from the burden of subsistence farming they discovered wants and desires at a scale that pulled entire markets into existence. Rao suggests the same might happen when people are freed from “subsistence cognition”. Rao concludes: What gets missed is the possibility that we are witnessing the beginning of a transition as profound as the one from an overwhelmingly agricultural society to an industrial and commercial one. From a world where 80% of human brains are engaged in subsistence cognition to one in which 95% are engaged in satisfying previously unfulfillable cognitive demands. The most important consequence of the agricultural revolution was not that farming became more efficient. It was that most people were liberated from the necessity of farming, allowing entirely new kinds of economic and cultural life to emerge. Likewise, the most important consequence of the AI revolution may not be that intelligence production becomes more efficient. It may be that most minds, human and eventually artificial, are liberated from the necessity of producing intelligence merely to survive, and become free to discover what they actually want to do with it. The cognitive mercantilist sees a world threatened by an oversupply of intelligence and an impending surplus of useless minds. The AI Malthusian sees a world in which intelligence outruns the capacity of humanity to absorb it, leaving a growing population of economically redundant minds primed for revolutionary explosions. The cognitive capitalist sees the possibility of an almost limitless expansion of demand, and a world full of minds whose potential has barely begun to be explored. I do understand why people have difficulty imagining this future. Capitalism, particularly over the last half century, has increasingly given way to scale, as ever larger corporations mediate ever more economic transactions, and as the Internet has played an important role in making the market a global one. The England in which capitalism emerged, however, and especially the America in which it was perfected, was much different. Alexis de Tocqueville, in Democracy in America , famously wrote: The political associations which exist in the United States are only a single feature in the midst of the immense assemblage of associations in that country. Americans of all ages, all conditions, and all dispositions, constantly form associations. They have not only commercial and manufacturing companies, in which all take part, but associations of a thousand other kinds — religious, moral, serious, futile, extensive, or restricted, enormous or diminutive. The Americans make associations to give entertainments, to found establishments for education, to build inns, to construct churches, to diffuse books, to send missionaries to the antipodes; and in this manner they found hospitals, prisons, and schools. If it be proposed to advance some truth, or to foster some feeling by the encouragement of a great example, they form a society. Wherever, at the head of some new undertaking, you see the government in France, or a man of rank in England, in the United States you will be sure to find an association. I met with several kinds of associations in America, of which I confess I had no previous notion; and I have often admired the extreme skill with which the inhabitants of the United States succeed in proposing a common object to the exertions of a great many men, and in getting them voluntarily to pursue it. I have since travelled over England, whence the Americans have taken some of their laws and many of their customs; and it seemed to me that the principle of association was by no means so constantly or so adroitly used in that country. The English often perform great things singly; whereas the Americans form associations for the smallest undertakings. It is evident that the former people consider association as a powerful means of action, but the latter seem to regard it as the only means they have of acting. My life is defined by the Internet: I witnessed and participated in its evolution, and maybe in some small way contributed to it; I certainly owe my livelihood and success to it. I bemoan its passing. And yet, in some ways, I am optimistic about what comes next. When it comes to economics, or social interaction, or now software, the only viable approach is to become smaller; smallness at scale, though, means there is not one global market, but infinite local ones. “Local” might mean geographic, or it might mean a private network; what many markets definitely means is many opportunities, because economies of scale matter less than the ability to meet new needs that only arise at the level of the association. This is where I come back to de Tocqueville: the Internet is an American invention, and has long been defined by American mores; the individuals who thrive on the Internet, no matter their nationality, come across as very American in their iconoclasm and cringe. Beyond the individual, however, there is the association, and amongst the association there is not a market, but infinite markets. Markets are just another word for needs and desires and the mechanisms of meeting and fulfilling them; the end of the Internet may mean the possibilities for creation are only just beginning. Presented by When I write about tech companies, I am often writing about businesses that depend on Stripe , because Stripe provides the foundation for Internet-native companies — including Stratechery. Consider Substack: building Stratechery was complicated and involved gluing together disparate pieces; Substack made it easy by delivering an integrated offering. One piece that was the same, however, was Stripe: Substack creators are also Stripe customers, as are creators on many other platforms. Zoom out and the pattern repeats. Fintechs can build entire business models around card programs without ever having to build the issuing and processing infrastructure underneath it. AI companies can charge for what customers actually use, which is particularly important when usage also drives costs, without having to build the systems to track and bill for it themselves. No business succeeds because of Stripe. Stripe doesn’t answer the hard questions about what to build or whether anyone will actually use it. But it does make more things practical to try. Talking Points Memo , Andrew Sullivan and Boing Boing launched in 2000 Instapundit , Little Green Footballs , and Dooce launched in 2001 Matt Yglesias , Daily Kos , Eschaton , The Volokh Conspiracy , Power Line , Gizmodo , Gawker , paidContent , and Daring Fireball launched in 2002 Ezra Klein , Marginal Revolution , ReadWriteWeb , and AVC launched in 2003 Compromised intermediate CA private keys (although this was an example of sheer incompetence, not AI hacking) Major new vulnerabilities in Node.js, GitLab, Java, and Cisco products Breaches in LiteLLM’s supply chain and Railway developer platform

0 views
Stratechery 3 weeks ago

2026.43: A New Era

Welcome back to This Week in Stratechery! This week marks a new era for Stratechery, which is now presented by Stripe . I’m excited about what this partnership can unlock, starting with in-person on-video episodes of Sharp Tech. Check out the YouTube video above (or watch it on Stratechery ); and, for subscribers, look forward to a bonus episode showing up in your subscriber feed every Sunday morning (add your feed on sharptech.fm ). Here were a few of our favorite topics this week: Presented by It’s kind of weird to be the Internet’s number one fan of ads. What makes it even weirder is that I have been the Internet’s number one fan of ads without actually having ads on Stratechery . That second part is changing: I’m super excited to announce that Stripe is going to be the presenting sponsor of Stratechery. And, at the end of the day, for me it couldn’t be any other company. One of my hangups in terms of advertising is that I take my ethics super seriously, and ultimately, if a company is helping pay the bills, it’s hard to be truly objective about them. What I realized, however, is that it’s hard for me to be objective about Stripe anyway.  I wanted to be a professional blogger long before I started Stratechery; however, I thought I had missed my chance, particularly as it became clear that advertising was not a long term business model for a one-person site. However, when Stripe launched with support for recurring billing, I realized there was a better model: subscriptions. 13 years later, and I’m still a happy customer. Or, to put it another way, I’m already conflicted because I love Stripe.  I’m not the only one: more than 5 million businesses use Stripe, accounting for 1.7% of global GDP. Moreover, a lot of those businesses are Internet native and accretive to the world’s economy, which is to say that Stripe is literally making the world a better place. That’s an easy company to get behind, especially since they have been behind Stratechery from the very beginning. — Ben Thompson The Best iPad; The Worst Phone. John Gruber wrote a great review of the iPhone Duo, which we broke down on Dithering . In short, it’s an iPad that fits in your pocket, and it’s awesome at that; just don’t expect a good phone. Plus, on Monday we discussed Apple’s surprising partnership with Meta for edge AI, a topic I also covered in Tuesday’s Update . — Ben Thompson The SS Sarandos in a Storm. The past week has seen a flood of bad news for Netflix: layoffs, executive departures, another disappointing round of earnings, and another hit to a share price that has taken a 40% hit over the past 12 months. Wednesday’s Update provides a comprehensive state of the union , and we doubled back on the topic for a fun conversation on this week’s Sharp Tech , where we divided the company’s problems into areas that can be fixed (the content) and what may be much trickier (growing the business in a world where Meta and YouTube are primary engagement competition). — Andrew Sharp You Can See Everything, But Can’t Trust Anyone. Allow me to be the 1,000th person on the internet to recommend Nathan Fielder’s new Elizabeth Holmes documentary, You Can See Everything . After declaring on Sharp Tech last week that I wasn’t interested in seeing it, I did in fact check it out, and wound up loving it. I wrote about my experience on Sharp Text , with 5 spoiler-free notes for anyone still on the fence, and 5 more reactions for anyone who’s seen it. Compulsively watchable (and worth seeing in a theater), I think it may have also been the most effective document of our modern media environment I’ve ever seen. — AS Do You Trust Apple? — Apple has a lot of trust in terms of delivering integrated products and preserving privacy; the real trust question is if you think Siri AI will justify investing more in their ecosystem. Muse Spark Open, Meta + Apple, Thin and Thick — Meta has a new open source model, and to everyone’s surprise, they are are partnering with Apple to make sure you can run it locally. Netflix Earnings, Business or Franchise, Disney’s Deal — Netflix has a solid business, which is exactly the problem: the company increasingly seems like a business and not a franchise, at least in terms of growth. An Interview with Craig Moffett About SpaceX and Bubbles — An interview with Craig Moffett about SpaceX, from launch to broadband to mobile, plus how he is applying the lessons of the dot-com bubble to today. You Can See Everything (And It’s a Great Time) — Nathan Fielder’s movie lives up to the hype and captures the modern world better than any art in recent memory. Apple + Meta AI The Duo Review The Birth & Rise of India’s IT Outsourcing Industry New Export Controls in Multiple Sectors; Q3 GDP and the Fifth Plenum; Europe Presses Forward; OpenAI and Nvidia Chips Aday Mara and the OKC Woodchipper, An Emergency Celtics Panic Watch, Silver Appreciation(???) and Laker Dreams Netflix Angst, The End of the Internet, Q&A on Meta’s VR Glasses, Anthropic Alignment, You Can See Everything

0 views