(Preview) OpenAI’s Enterprise Pivot, The Rise of Agents and Bubble Counterpoints, Nvidia Changes Its Inference Story
- OpenAI’s reported reset toward coding and business users is less a retreat than a correction toward the market most willing to pay for productivity. Andrew notes that OpenAI has spent the last 3 years pursuing consumer subscriptions. Ben Thompson’s Dropbox analogy carries the logic: an exceptional consumer product eventually had to rebuild for enterprise permissions and authentication because “consumers don’t pay for productivity apps,” while employers readily subscribe when software makes salaried workers more productive.
- The immediate strategic danger is Anthropic, which Ben says is “blowing up in the enterprise.” He cites an apparent run-rate jump “from $14 billion in January to $19 billion” now, while acknowledging that private-company figures are difficult to validate. Ben also says OpenAI has much more compute than Anthropic, with that compute serving consumers; Andrew sees a direct 12-to-24-month opportunity, while Ben warns enterprises could standardize on Anthropic and lock OpenAI out.
- The shift does not mean OpenAI is abandoning consumers, despite the dramatic reactions to its all-hands message about avoiding “side quests.” Sora is an example of a side project that could be deprioritized, but Sam Altman explicitly denied that the hardware effort was shutting down: “Quite the opposite. I think you will love what the team is building.”
- ChatGPT’s enormous consumer reach is simultaneously OpenAI’s strategic advantage and “biggest problem,” because supporting that audience consumes vast compute without a proven matching revenue engine. Advertising could unlock the larger consumer market, but OpenAI would be building an ads system from scratch while Google and Meta already possess mature infrastructure enhanced by AI; enterprise subscriptions offer a clearer near-term path to cash.
- Enterprise AI adoption is partly an organizational-design problem because chatbots require employees to choose to use them, and many workers “are just there to collect a check.” Top-down incentives can produce resentment or superficial compliance: Microsoft’s KPI to integrate Copilot yielded it “everywhere,” including places where Ben says it was unwanted or worked poorly.
- Ben’s counterpoint to the AI-bubble case is that old GPUs appear more economically durable than depreciation skeptics assumed, with their prices actually rising. Major platforms are spending roughly around projected free cash flow rather than becoming broadly overleveraged; debt can rationally match upfront infrastructure costs with returns earned over time, while Oracle is the notable aggressive borrower backed by sharply expanding committed business.
- Apparent Anthropic share gains should be discounted for selection bias, particularly when the evidence comes from Ramp customers. Ramp disproportionately serves startups and highly technical Silicon Valley companies—the same cohort most likely to adopt the “hot Silicon Valley company”—whereas general Fortune 500 companies, Ben says, know OpenAI; after company pushback, he gives OpenAI “the benefit of the doubt.”
1. OpenAI is rediscovering the business model of productivity software
Andrew frames the reported strategy shift through Fidji Simo’s warning that OpenAI could not afford to be “distracted by side quests.” Leadership was examining what to deprioritize so the company could “nail productivity in general, and particularly productivity on the business front,” with coding and business users becoming the center of gravity. Andrew notes that OpenAI has spent the last 3 years pursuing consumer subscriptions.
Ben’s historical specimen is Dropbox, once a mind-blowing “USB drive in the sky.” At English schools around Taipei, he used Dropbox, Mac Minis, AppleScript, and nightly jobs to synchronize Keynote curricula across classrooms—an improvised system whose real engine was Dropbox’s effortless promise that a file placed in one folder would appear everywhere.
Yet Dropbox’s consumer success did not produce the right economics. It spent years rewriting its application around enterprise permissions, authentication, and administration because, outside enthusiasts like Ben, “consumers don’t pay for productivity apps”; enterprises both pay for them and demand the security, support, and updates required for deployment.
Ben traces the template to Microsoft’s subscription transition under Steve Ballmer around the late 1990s or circa 2000. For a “logical spreadsheet-driven CTO,” recurring payment matches recurring utility: employees remain productive, support stays available, security is maintained, and updates arrive immediately. Consumers, by contrast, usually prefer paying with attention through advertising.
2. Human incentives, not model capability, may cap chatbot adoption
Ben’s qualification to the productivity thesis is that chatbots require active employee participation. Some workers want to improve their jobs, but many simply “punch in and punch out” and may resent demands to become more productive; Andrew cites finance friends with quarterly meetings where they must explain how they are using AI in their jobs each week.
The organizational problem is getting hundreds or millions of employees “rowing, broadly speaking, in the same direction.” People optimize toward local KPIs, so Microsoft’s KPI to integrate Copilot predictably produced Copilot everywhere—including places where Ben says it was unwanted and did not work particularly well.
His model of corporate management is a zigzag: incentives initially move an organization toward its goal, then overshoot as employees maximize the metric, forcing another reorganization and course correction. Ben calls this the “nerd fallacy”: computers are difficult to understand, but once mastered execute even flawed instructions exactly; people and organizational headwinds are much harder to manage.
3. Current AI spending does not yet resemble broad overleverage
Ben addresses the depreciation argument behind some bubble calls: critics assumed GPUs would be useful for only a couple of years, making reported losses artificially low. “Turns out that doesn’t appear to be the case,” he says, because prices for old GPUs are rising and companies are actually making more money from hardware that has already been depreciated.
The major platforms’ CapEx broadly tracks projected free cash flow: Meta and Google are near that number, Microsoft is somewhat below it, and Amazon is somewhat above it. Ben’s pushback to alarm over Amazon spending more than free cash flow is straightforward: capital investment consumes cash upfront and earns returns over time, while debt supplies cash upfront and is repaid over time.
In Ben’s framing, debt is “not a bad thing” but an underused and tax-advantaged tool for technology companies. Oracle is the conspicuous company taking on substantial debt, yet its earnings were “incredible,” and its RPO—committed future business—rose by “like, $50 billion or something like that.” Andrew’s test for a truly bubbly build-out was that all the major companies would be overlevered, not merely spending near internally generated cash.
4. The enterprise focus can coexist with OpenAI’s consumer ambition
Ben cautions against reading the all-hands report as a wholesale consumer exit. The Sora app was a memorable experiment that “didn’t ultimately matter to the bottom line,” making it a plausible side quest to cut; hardware appears protected after Altman said, “We are not shutting it down. Quite the opposite.”
Listener Adrian supplies the strongest pushback: OpenAI’s consumer platform could create a larger flywheel as general-purpose models, smartphones, watches, glasses, and other interaction points improve. His internet analogy is that early government and academic adoption eventually gave way to enormous everyday value, with scale enabling targeted advertising that benefits advertisers and users.
Ben concedes that consumer is generally the larger market and that ChatGPT’s scale should, in theory, support an ad-funded business. The obstacle is execution: “Oh my God, building ads is hard.” Google and Meta already own the machinery and are improving it with AI, whereas OpenAI must build from scratch while supporting the compute-intensive consumer audience. That scale also means, Ben says, OpenAI has much more compute than Anthropic, with that compute serving consumers.
His counterexample is Microsoft in the 1980s and 1990s: enterprise dominance created the Windows flywheel, effectively delivering the consumer market “for free” and even seeding Microsoft’s later gaming position. Apple showed the opposite risk, nearly failing while charging consumers a premium before enough buyers valued a differentiated computer.
5. Anthropic creates urgency, but its apparent lead is difficult to measure
The strategic pressure Ben identifies is Anthropic, whose enterprise growth he describes as an exponential curve: “they’re growing, they’re growing, they’re growing, and then holy crap, they’re growing.” He references a roughly $14 billion run rate in January becoming about $19 billion now, while treating the private-company numbers as inherently uncertain.
Andrew argues OpenAI has direct line of sight to what coding and enterprise products could accomplish over the next 12 to 24 months, especially given Codex’s performance. Ben sees the downside condition clearly: if OpenAI waits, customers could standardize on Anthropic and effectively lock it out of the most reliable subscription market.
Ben nevertheless retreats from treating Ramp’s spending data as definitive evidence of Anthropic’s lead. Ramp over-indexes toward startups and technical Silicon Valley customers, who also over-index toward Anthropic; general Fortune 500 companies, Ben says, know OpenAI instead. After “very strong pushback” from OpenAI, this is one case where he grants the company the benefit of the doubt.
Full transcript
Hello, and welcome to a free preview of Sharp Tech. Hello, and welcome back to another episode of Sharp Tech. I'm Andrew Sharp, and on the other line, a day early this week, Ben Thompson. Ben, how are you doing?
I'm doing okay. Every time I'm sitting around writing multiple times a week, I figure, “What am I going to write about?” And then, at the moment I'm about to take a couple of days off, I have 57 gazillion things that I want to write about. We're going to have to cover a lot of bases on here. Hopefully people will listen.
There are 57 gazillion things that we could cover in the next 90 minutes on this episode, so we're not going to be able to get to everything that's worth covering. In any event, is this your final piece of content before your spring vacation?
This is, and it's going to be a weird vacation. I'm not taking the whole next week off.
Mm-hmm.
I'm going to publish a couple of days, take 1 day off this week, and 1 day off the following week. I'm really doing a terrible job of taking time for myself, but it is what it is.
Well, we'll work on it next year. Next year you're going to host me for March Madness in your basement with your 10 TVs and wonderful couch. I'm very upset that you're leaving right now.
No, I think it's going to be a good year this year, too. Lots of good teams, lots of good NBA prospects.
It's very exciting.
I'm actually kind of annoyed that—
I know.
I'm about to leave.
I'm dialed in for the next couple of weeks. In any event, we'll begin. You planted your flag with an article on Monday, “We Are Not in a Bubble,” which we had talked about in the past, and you'd been leaning in that direction. We've had some listeners who were leaning that direction pretty aggressively. We got a number of questions related to that article, but I want to start with a story that ran Monday afternoon, after your article Monday morning, in The Wall Street Journal—
Yeah.
—that I think dovetails—
A good example of a point that I have not covered at Stratechery that I want to— But yes.
Yes. Well, that's what the podcast is for. The headline from The Wall Street Journal: “OpenAI to Cut Back on Side Projects in Push to, Quote, ‘Nail Core Business.’” The Wall Street Journal writes:
“OpenAI's top executives are finalizing plans for a major strategy shift to refocus the company around coding and business users, recognizing that a do-everything-all-at-once strategy has put them on the defensive. Fidji Simo, OpenAI's CEO of Applications, previewed the changes to employees in an all-hands meeting, telling them that top leaders, including CEO Sam Altman and Chief Research Officer Mark Chen, were actively looking at which areas to deprioritize. ‘We cannot miss this moment because we are distracted by side quests,’ Simo told staff last week. ‘We really have to nail productivity in general, and particularly productivity on the business front.’”
So, Ben, how do you understand that news in the context of AI in 2026 and OpenAI specifically in 2026?
It's funny because I think the takeaway is one of the oldest Stratechery axioms that I wrote about for years. It took me a while to come to it— a year or two. I think the company that I always think about is Dropbox, where Dropbox was an incredible product.
Yeah.
Maybe you were too young to remember this, but back in the day, I don't know if you were still running around in elementary school with your football jerseys or whatever.
What I can say is I've come to love Dropbox over the last 4 years working at Stratechery.
Back when Dropbox was a thing, I remember I sort of blew everyone's mind at business school, where people were still using USB thumb drives to give files around. I was at some stupid networking event, and I needed something like a résumé or whatever it might be. I just went to a computer, logged into Dropbox, and got what I needed. It was there. I might have even changed it slightly to be tuned to whoever I was giving it to, or whatever.
But this idea that it's like a USB drive in the sky—this technology that was amazing and solved all these—
Like—
—you can have your applications anywhere, at any time, whenever you need them.
And Dropbox had come out, I think, fairly well before then, but I was a super... Actually, it had been out for several years at this point, I believe. And I actually used it at this school that I was working at for some English schools in Taiwan, and they had these massive whiteboards where you would write out these sentences for drilling and going over as a class and all. Literally, there would be like 10 minutes where people would just sit there and the teacher would be writing out all this sort of thing on the board.
Scrawling out the lesson on the whiteboard? Okay.
Yeah. I took this whole curriculum and put it into Keynote, but then put Mac minis in every classroom. There were multiple locations, so it was spread out all over Taipei and a couple of locations outside of that. The question was, how do you then actually keep all those in sync?
Yeah.
There was a master Mac mini, and I wrote a couple of AppleScripts that would basically push stuff out, so you could do stuff on the Mac mini there. It would push it into a dedicated Dropbox folder that synced with all these Mac minis all over, and then each Mac mini had a cron job that ran at night. It would take everything in the Dropbox folder and push it into the actual folder that was used for teaching.
Wow.
The point of this was that you didn't want teachers to be able to randomly push back and undo everything. The long and short of it is that I built this, quote-unquote, “syncing engine” that was not me at all. It was just Dropbox, which was this technology that was amazing and solved all these—
Don't sell yourself short. It was a little bit you and a lot of Dropbox, saving 10 minutes in classrooms all over Taipei and beyond.
The point, though, was more than that because you had to do it multiple times a lesson. But the big thing was that it was so easy to use.
Mm-hmm.
It was a folder in the sky. You put stuff in the folder, and if it was in that folder, it would be everywhere you were signed into Dropbox.
Yeah.
That could be other computers, or it could be logging in through a web browser, all that sort of thing. From very early on, I have always been on the highest Dropbox tier. I've kept all my files there. I haven't tested this in a while. I used to test it fairly regularly to make sure it worked.
I should be able to lose my computer or have it fall to the bottom of a lake or, in San Francisco, get my car broken into and have it stolen, or whatever it might be. I should be able to walk into a Best Buy or an Apple Store, buy another computer, or even go into an—
Internet cafés?
Internet cafés. Yeah. I should be able to do my work.
Yeah.
I was pretty disciplined about that. Again, I've probably slacked off around making sure that all still works, but I'm pretty confident it still will. It's gotten more and more challenging for security reasons because everything requires two-factor authentication. You need a core piece that you can sign into so that you can get all the stuff you need to sign into everything else.
That's true. If your backpack gets stolen, it might be a problem logging into Dropbox.
Right. What if you don't have your phone? The biggie is, what if you don't have your phone?
Exactly. If your phone is in your backpack along with the laptop—
Right. This gets into the fact that I carry backup phones, but even if you lose your backup phone, then what do you do?
We're all over the country, and we're working from the same base, working from the same documents that are edited instantaneously any time any one of us does anything to them, all of which is made possible by Dropbox.
That's all sort of table stakes. Actually, we use a hodgepodge of stuff. Our documents are with—
There's SharePoint, yeah.
But anyhow, it's commonplace now. It's totally normal, table stakes. You expect stuff to sync. Apple has it built in now with iCloud. But back 17 or 18 years ago, it was a new thing, and it was awesome. Dropbox really appealed to and blew up in the consumer space.
They had a great viral bit, which was that if you got someone else to sign up, then you got more storage, which was—
But what happened to Dropbox is that they tried to build these consumer features, like some of this photo stuff. They had this whole Carousel thing and all these other features. What they ultimately had to do was sit down and completely rewrite their application and basically be dead in the water for a couple of years. They had to rewrite it so it would be suitable for enterprise.
That meant all these permissions and tying into all the enterprise authentication standards, all these bits and pieces. The reason is that while people like me—nerds—were happy to pay for Dropbox, consumers don't pay for apps.
How do they pay the market?
They don't pay for productivity apps.
Yeah.
Productivity apps have been something I've cared about for ages and ages. It's the reason I liked the Mac, even back when the Mac sucked, because it blew people's minds when I went to Microsoft and had to use Windows. People were like, "What do you miss about a Mac?" I'm like, "I miss third-party applications." They're like, "What? I thought the whole thing with Windows was that it had all the third-party applications."
Mm-hmm.
Well, Windows had all these applications that enterprises would write, like line-of-business apps, but those went away, by and large, to be replaced by SaaS apps. What the Mac always had was this whole ecosystem of small developers who would make these bespoke apps that were super useful, and those just didn't exist, or didn't exist at a similar quality, on Windows. I love all that stuff. I love productivity apps.
Yeah.
I wrote a lot in the App Store about how Apple wasn't enabling a business model for productivity apps. For a long time, the App Store was, you bought an app once and that was it. I wrote tons of articles saying, "You need to support subscriptions. You need to support upgrade pricing." They still don't support upgrade pricing, which irks me, but at least they support subscriptions for apps now.
Mm-hmm.
One of my earliest articles was explaining why subscription pricing is good for productivity apps. You're getting ongoing utility from it. You want it to get better. That matches up your payment with what you get. That's all right, and consumers hate it.
Yeah.
Consumers increasingly will pay subscription fees for apps, but they don't like it. It has to be really important to them, and they're still going to complain all along the way. Meanwhile, the enterprise has been paying. Microsoft shifted to subscription pricing ages ago, around 2000, maybe in the late '90s. This was the thing that Steve Ballmer executed that was actually really impressive. He obviously gets lots of criticism and complaints.
Yeah.
But shifting Microsoft to being a subscription business was genuinely groundbreaking. It aligned the delivery of value with the capture of value in a way that made it so, if you're a logical, spreadsheet-driven CTO in an enterprise, you're like, "Yes, I'm getting value from this on an ongoing basis. I'm happy to pay for it on an ongoing basis. I have support lined up. There's a security story. There are going to be immediate updates." It's just a good fit for productivity subscriptions. It all goes together. But you have to be hyper-rational to appreciate how good it is.
Yeah.
Enterprises can be hyper-rational about that, and they're paying employees, so if they can make those employees more productive, it's well worth the price. Consumers aren't.
Right.
They just—
Well, OpenAI has been working with a subscription model, chasing the consumer market, for the last 3 years here. If you're going to be working with a subscription model, the target should be enterprises, and I think they're now clear on that point.
Right. So what happened in the consumer space is that consumers were fine with ads.
Mm-hmm.
They didn't really care about their data. Enterprises obviously care a lot about their data. They don't want an ad-supported model for the enterprise; they want to pay. Consumers pay, too. They just pay with their attention.
Yeah.
You capture that via advertising. So you have this great bifurcation where, on the enterprise side, you pay with subscriptions, and on the consumer side, you pay with ads. That's the business model that works, by and large.
It's weird. Sam Altman has been in Silicon Valley forever. A lot of people involved in OpenAI have been in tech for a long time, and it's always a reminder that this is why Stratechery has a business, because no one else thinks about this stuff as much as I do. They were convinced for ages that the theory of the case was—and I'm pretty sure Sam Altman told me this in an interview sometime; I can maybe go dig it up—that AI is such a boon and makes you so much more productive that we will—
Everyone in America—
—basically defeat gravity.
—is going to subscribe. Yeah.
That's right, and consumers will subscribe. I'm like, "No, they won't." They're just not going to. Yes, it's impressive how many people subscribe to ChatGPT. It's impressive how much revenue OpenAI makes from what was, by and large, consumer-focused subscriptions for a long time.
Mm-hmm.
But you're going to hit a ceiling, because the fact of the matter is that most people aren't going to pay, and most consumers don't want to be productive.
Right.
They want to be entertained. An aspect of work—even at work, people don't necessarily want to be more productive. This has always been a question, which we'll get to in a little bit, but I think there's been a ceiling on chatbots in the enterprise because chatbots entail people actively using them.
Mm-hmm.
Yes, some people love their jobs and want to do them better and be more productive. You're going to hit a ceiling as to how many people are like that.
Right.
So—
Some people just punch in and punch out, you know? Punch in at night.
If anything, they're annoyed at having to be more productive.
Yeah.
There are lots of interesting studies. There is—
Oh, and there are millions of people who are annoyed by the top-down edicts to integrate AI into their daily workflow. I've got some friends in finance who have quarterly meetings where they have to explain how they're using AI in their job every week. Not everyone loves that, let me tell you—anecdotally, at least.
It's interesting. I didn't even think about this until now, but this is something that you spent a lot of time on in business school, which seems like a waste. Everyone in tech hates business school and people from business school, and understandably so. I mostly hate them, too.
Mm-hmm.
There's so much about organizational design, KPIs, incentives, and how you compensate people. That's actually the hardest part of business. It's funny because business school gets looked down on as, "How much work are you actually doing?" Which is true.
Mm-hmm.
There's a lot of quote-unquote networking.
Some of the trips that the MBA students take.
Which generally entails a lot of drinking.
Yeah.
I feel compelled to make sure my kids know how to ski.
Kind of bullshit relative to law school.
Yes. Going on the ski trip is super important in business school.
Yeah.
Actually, this specific bit is the hardest part of business.
Mm-hmm.
It's a part that tech people, particularly people who are entrepreneurial and love computers and are always trying to figure out how to do new stuff on computers because it's fun and interesting, have no theory of mind for.
This bit being coordinating across entire organizations?
Incentivizing. How do you get a large organization of hundreds, thousands, or tens of thousands of people—
Mm-hmm.
—hundreds of thousands, and some companies like Amazon, millions—
Yeah.
—rowing, broadly speaking, in the same direction and doing the things—
Using the same tools.
—that you want done—
Yeah.
—when the vast majority of people in your organization are just there to collect a check because they—
Mm-hmm.
—need money? That's why you have organizations and why they reorganize every few years.
Or they change stuff around. Because what happens is you design an incentive structure and an organizational structure, and you set certain KPIs, and everyone ends up optimized to their local maxima. They will maximize their KPI, right? So in Microsoft, there’s clearly been a KPI about integrating Copilot.
Mm-hmm.
What happens is you get Copilot everywhere, in places you don’t want it, and it doesn’t work very well, but that’s what they’re incentivized to do, right?
Yeah.
And so Microsoft now, I think, just added another—or some sort of reorg is going on. It’s like you don’t approach the organizational goal directly. You approach it in a zigzag pattern—
Mm-hmm.
—where you’re kind of going in the right direction with your KPI and structure, and then you’re like, “Wait, we’re going way off course now. Time to reorg, time to re-incentivize.”
Right, and then, yeah.
And then you turn to the right, and you’re kind of going in the right direction, and then you go past it and you’re going somewhere else, and then you have to do it again. And it’s easy on the outside to look at this and to think, like, “These stupid business people. What are they doing?”
Mm-hmm.
Like, “Why can’t they just get it right the first time?” Because they’re dealing with people, and people are a lot more difficult to deal with than computers. And that is sort of the nerd fallacy, which we’ve talked about in the context of AI, AI replacing jobs, and all these sorts of things. To understand and deal with a computer is exceptionally difficult, and the number of people who can do it well is quite small.
Yeah.
But at the end of the day, if you can master it, the computer does what you tell it to. The thing about bugs is that there are a very small number of situations where the bug is not the fault of someone in the programming stack. There was a famous calculation bug in Intel processors, I think in the 1990s, where it would actually make the wrong calculation, and it was this big thing. Technically, it’s a bug because it was in the design of the actual chip itself, but by and large, a quote-unquote “bug” is you told the computer to do the wrong thing. You didn’t realize you were telling the computer to do the wrong thing, but the computer did exactly what you told it to—it’s just that your instructions were flawed.
Well, and you can easily identify a bug and address it.
Well, easily in quotes. But yes, you can at least in theory easily identify the bug and find it.
Well, it’s easier to identify a bug than to identify a cultural headwind that you’re dealing with across an organization, for instance.
Well, you’re dealing with people.
Yeah.
And dealing with and managing people in large organizations is exceptionally difficult. There’s a reason why there are entire disciplines, entire schools devoted to this question, and everyone looks at them and says, “Wow, you seem to do a really bad job because you seem to be producing graduates that are changing all the time and don’t know what they’re doing.” That’s because they’re trying to manage humans.
So you’re saying there’s a reason that people go to business school. I don’t know if I believe you, but—
Yeah, maybe not. They go to business school to get a job. But yes, no—this is a huge part of business school. I’ve talked about getting a lot of value out of the game theory classes and accounting, actually.
Uh-huh.
I was annoyed I had to take these accounting classes. Actually, number 1, accounting is super interesting. I really liked it. And number 2, it’s also been exceptionally valuable these days, right? One of the big “this is a bubble” arguments has been about the question of depreciation, right?
Mm-hmm.
Like, how many years are these GPUs being depreciated over? Actually, they’re only good for a couple of years. These companies’ losses are actually going to be much larger than they’re showing on their books, et cetera. Turns out that doesn’t appear to be the case. The price of old GPUs is going up.
Hmm.
So companies are actually making more money on old GPUs, including ones that are depreciated. But the question of depreciation and CapEx, all those sorts of things—accounting questions—and how do you actually represent the business. I don’t get into this too much, but there’s the question of cash flow versus your accounting, because especially with depreciation, that depreciation money goes out the door right away, and then you only account for it over time, so you need to look at cash flow.
Mm-hmm.
On the other hand, questions of debt, right? It’s funny because everyone is scandalized about Amazon projecting more CapEx than free cash flow.
Yeah.
And it’s like, “Well, how do they do that?” Because they issue debt. That’s the thing: for most normal companies, they actually carry a significant amount of debt precisely because debt matches capital investment. Capital investment—you put a lot of money in up front, and you reap the gains over time. What is debt? You get a lot of money up front, and you pay it back over time, right?
Yeah.
There’s a reason—debt is not a bad thing. It’s an incredibly useful and important tool, right? And actually, one of the funny things about tech is how little debt tech companies have traditionally carried, and this is actually a bad thing—not just because it matches up the cash flows, but because our tax code dramatically favors and gives you real benefits for using debt.
Mm-hmm.
So there’s actually a bit where all these companies have been significantly under-monetized by not having the correct capital structure, like the amount of debt they would carry. This is something I wrote about a decade ago, when Apple first started issuing debt.
Yeah.
It’s like, “Why is Apple issuing debt?” You think about it from a consumer mindset: wasn’t it good that Apple has hundreds of billions of dollars or whatever in the bank? It’s like, no, that’s actually—
No.
—that’s very suboptimal, right? So that’s the other thing about the fact that every company is spending up to its free cash flow—
Yeah.
—which basically, if you wanted to predict how much the tech companies are going to spend on CapEx this year, you could just look at what their projected free cash flow was, and that’s basically what Meta’s spending.
They’re hitting that number, right.
That’s what Google’s spending. Microsoft is under it a bit, which we can talk about. Amazon is over it a bit, but that’s the ballpark. But the reality is they can all go much further. They could spend way more if they start getting into debt, like Oracle.
If it got really bubbly, they would all be overlevered on the infrastructure build-out.
Right. The one that’s really in debt or taking on debt—
Or Oracle.
—is Oracle, right?
Yeah.
But Oracle’s earnings were incredible this year. And their revenue RPO—revenue purchase obligation, basically the amount of committed business they have, so you have to put it on your books as a liability because it’s like, “We’ve agreed to serve this”—increased by $50 billion or something like that. The numbers are nuts. Anyhow, I’m way off base.
No, we’re going back to OpenAI now.
Oh, we’re going? All right. Thank you for throwing me a rope as I’ve drowned in this as well.
No, but it’s been terrific. We got a free business school class there at the top of the show.
So OpenAI’s problem is, yeah, they were predicated on a business model that history says doesn’t work.
Well, hold on. Let me read this email from Adrian because I am talking to the author of The Accidental Consumer Tech Company here about OpenAI, and Adrian says:
“OpenAI should not pivot to coding and enterprise. This is a terrible pivot. Unless you think Google can’t be beaten in the consumer market, the scale that OpenAI could achieve with a consumer platform creates a flywheel that you just can’t achieve in the smaller, more competitive enterprise market. Professional applications of AI are the most interesting right now, but that is bound to change as, 1, general-purpose AI models improve; 2, smartphones, watches, glasses, jewelry, et cetera, improve as AI interaction points; and 3, consumer adoption increases.
“The internet’s early adopters were government and academia, but so much of the value created on the internet has ultimately come from everyday consumer use cases. The business model of the internet is ads, which works because the massive audience makes for great ad targeting, which means ads actually make advertisers money and serve products that are interesting to users.
Thanks for teaching me this, Ben.
You're welcome, Adrian.
Adding value everywhere is better than adding value for large enterprises. Am I underrating the value of enterprise solutions? If so, tell me how I'm wrong. What do you think about that?
A couple things. Number 1, the OpenAI rumor, all-hands meeting, whatever, didn't say they're exiting the consumer business, right?
Yeah.
We know they have side quests they can get rid of, like Sora, for example.
Mm-hmm.
Like the Sora app. We spent a lot of time on that. We've been chastised, I think, to a certain extent.
It was a fun couple of weeks, but it didn't ultimately matter to the bottom line.
Right. There are a number of other things. From what I understand, another thing to potentially point to is the hardware question.
Mm-hmm.
On the hardware question, Sam said, "No, we're not getting out of it," and I'm pretty sure they're not getting out of that.
He said, "We are not shutting it down. Quite the opposite. I think you will love what the team is building."
And so that's probably the tell. Are we giving up the consumer market? I don't think they are.
Mm-hmm.
Number 1, let's not overread what they said.
Right. I think some of the reactions online, not just Adrian, were a bit too dramatic. I didn't come away thinking OpenAI is just going to punt on the consumer space altogether. It makes sense as far as focus going forward. There's a massive opportunity in the enterprise space right now on a subscription basis. It makes sense to channel more resources in that direction, particularly given how performant Codex has been, which we discussed last week.
Right.
All of this is very rational.
Well, Anthropic is threatening to basically just win the whole thing. That's what's really happening here: Anthropic is blowing up in the enterprise. They have been for a while. Their growth over the last 2 years looks like an exponential curve: they're growing, they're growing, they're growing, and then, holy crap, they're growing.
Yeah.
From a $14 billion run rate in January to $19 billion now, something crazy like that. Just a dramatic ramp-up. And the issue is, enterprises do pay for productivity.
Yeah.
If AI is indeed a massive productivity enhancer—such a massive productivity enhancer that it not only makes employees more productive but potentially replaces employees—the size of that market is probably larger than Andrew's giving it credit for.
Mm-hmm.
And that may be a sufficient flywheel to then go into the consumer market. There is an analogy for this, which is Microsoft in the '80s and '90s.
Mm-hmm.
Microsoft was an enterprise company. They got the Windows flywheel going in enterprise and then basically won the consumer market for free.
Yeah.
Microsoft itself forgot this lesson, right? When they were trying to do things like the Zune, it was like, what? They were never a consumer company. They won the consumer market by virtue of dominating the enterprise, and then they sort of got consumer for free. They still have a hangover from that. Why is Microsoft still dominant in terms of gaming? That's downstream from winning enterprise in the '80s and '90s, right?
They've made investments there, but they had critical mass at the moment when gaming blew up and became a big thing. Everything's built around Windows—the components, the hardware, and all these bits and pieces. They still maintain a consumer presence, but they've always been an enterprise company that sort of got that for free.
Right.
Whereas Apple was a consumer company that was in the wilderness and almost went bankrupt because they were trying to sell to consumers at a time when consumers weren't willing to pay what Apple was asking.
They were charging a premium for the first 20 or 30 years.
They were charging a premium, and there weren't that many people in the consumer space who cared enough to get a computer and cared enough to buy a differentiated computer. It was just easier to get what you got at work, which, by the way, was much cheaper because you had this flywheel going. The computers were much less expensive and more performant, and just everything about the Apple thing was worse.
So, just to push back on Adrian, we do have an example in tech history of the more important flywheel being enterprise, which lets you win consumer.
Mm-hmm.
And when we haven't fully fleshed out the business model, we know enterprise will just pay. They'll pay for what they're getting, and that's super powerful.
Yeah, and we could lose it if we don't seize the opportunity over the next 1 or 2 years here.
Right. And so this gets to the accidental consumer tech company bit. My thesis has been that OpenAI is large, ChatGPT is massive, so you have to get into a consumer business model, which is ads.
Mm-hmm.
There's a bit where the sheer scale of ChatGPT is actually OpenAI's biggest problem. They have to support so many people, and they have so much more compute than Anthropic. But that compute is all going to serve all these consumers. So we have to build an ad business. Oh my God, building ads is hard. Google already has it built, Meta already has it built, and they're accelerating their offering because they're applying AI to it, while we're starting from scratch.
Mm-hmm.
In theory, yes, they should have a huge ad-supported consumer business, and maybe they add devices to it and all this sort of thing, but it's a long and difficult row to hoe. In theory, that is the better flywheel. The consumer market is larger than the enterprise market, generally speaking. But they need to actually make money to pay for all this sort of stuff. And making money is easier in enterprise, and it's probably a pretty large market—larger than Andrew's giving it credit for.
Well, there's a direct line of sight in terms of what OpenAI could do in the next 12 to 24 months if they seize the opportunity and go after it aggressively.
And there's a bit where they just get locked out because everyone's sort of standardized on Anthropic, right? That's—
Exactly. That's the risk.
Yeah.
That's the risk that's animating a lot of what's happening here. And it's hard because all of these are private companies, so you'll see these charts about Claude eating into the enterprise space, and who knows what exactly is real.
Yeah. I posted that chart from Ramp a couple of weeks ago. I got some very strong pushback from OpenAI that they don't believe it's close to being accurate.
Interesting, yeah.
The thing with Ramp is, it's used by a lot of startups and very techy companies. These Silicon Valley products are used first and foremost by other Silicon Valley products. Ramp, by all accounts, is doing extremely well and having a very strong push in enterprise. Most of your big enterprises are still with American Express, right? Or—
Yeah.
And in the largest companies—
The sorts of businesses that are most likely to adopt the cool new technology are the startups that are using Ramp.
Right. Those are the exact same companies that are more likely to be using the hot Silicon Valley company—
Right.
—which is Anthropic. Whereas your general Fortune 500 companies know OpenAI.
And so that's probably a very unreliable chart. I put a sentence when I quoted that saying, "This might not be a reliable chart." I've gotten lots of pushback from companies, and I've gotten very good at ignoring it. But I do think this is one where I'm going to give OpenAI the benefit of the doubt.
All right, and that is the end of the free preview. If you'd like to hear more from Ben and I, there are links to subscribe in the show notes, or you can also go to sharptech.fm. Either option will get you access to a personalized feed that has all the shows we do every week, plus lots more great content from Stratechery and the Stratechery Plus bundle. Check it out, and if you've got feedback, please email us at email@sharptech.fm.