“We’re doing well more revenue than [$13 billion]. Brad, if you want to sell your shares, I’ll find you a buyer. Enough. There’s a lot of people who want to buy OpenAI shares.” —Sam Altman, in response to a simple question from Brad Gerstner, one of his own investors, who asked how a company with $13 billion in revenue could have $1.4 trillion in spending commitments. Altman went on to admit that he is “taking a forward bet” that revenue will continue to grow, but that “if we don’t have the compute, we won’t be able to generate the revenue.”

Yes, that logic is a bit circular. Which is a great segue to our main event…

Between the hundreds of billions of dollars that A.I. companies are spending on data centers, their multitrillion-dollar valuation growth, and the nagging question of whether they’ll ever generate enough revenue to justify those investments, it’s easy to conclude that Silicon Valley is entering bubble territory. And certainly, there are prevalent themes today that have accompanied most historical bubbles—an irresistible narrative, valuations that are divorced from fundamentals, high capex, etcetera.

That’s why Tejas Dessai, the analyst leading thematic research at Global X, believes this was a “critical” quarter to determine the health of the A.I. trade—a bull market that’s been raging since 2023, back when Nvidia was valued at just $1 trillion. Last week, of course, it became the first $5 trillion company in history. Investors, Dessai told me, were looking for two things above all else: evidence that the trade remains intact, and a commitment to revenue generation. And on those two points, the four major hyperscalers—Microsoft, Amazon, Google, and Meta—all delivered this past quarter, with each significantly boosting their capital expenditure forecasts. Meta raised the bottom of its capex range to $70 billion; Amazon went from $118 billion to $125 billion; and Alphabet went from $75 billion to $91 billion. Microsoft, which told investors a few months ago that it would start slowing capex, turned up the wick instead. All told, those four companies are set to reach roughly $380 billion in capex this year, a number that will likely rise in 2026 as the need for compute increases. (Last time we talked about the bubble, in August, these companies were committed to $325 billion in capex by year’s end.) Obviously, the revenue side of the equation is lagging far behind the scale of the infrastructure investments being made. Still, demand exists, even if it’s not always clear where it’s coming from. Gil Luria, head of technology research at D.A. Davidson, told me that some demand for compute comes from “real companies” adopting enterprise A.I. tools, and that much of the rest comes from consumer applications like ChatGPT, which aren’t profitable. Everyone is just betting that both demand and revenue will continue to grow and eventually outpace build-out costs. In the meantime, the hyperscalers are diverting cash from more successful businesses, which means their collective free cashflow numbers are also steadily dwindling. This is perhaps most visible with Amazon, whose free cashflow sagged to $14.8 billion for the trailing 12 months, from $47.7 billion for the same period the year before. Wall Street’s response has been mixed. Amazon and Google, which both sell cloud services to other A.I. companies, surged about 10 percent and 5 percent post-earnings, respectively. But shares of Meta and Microsoft both dropped. The decline was particularly notable at Meta—the only hyperscaler that is not a cloud company, and so has a less obvious road to R.O.I.—which tumbled about 10 percent after releasing its numbers. According to Dessai, it was a healthy moment of the market performing a “self-check,” demanding more information before rewarding the stock.

To uncover more of the story for myself, I dug into the S.E.C. filings for each company to assess how much money they’re actually raking in from the A.I. boom. And here, too, the results were mixed.

Amazon posted a huge jump in net income year over year, from $15.3 billion to $21.2 billion in the third quarter. However, that increase included $9.5 billion in pretax gains from Amazon’s investment in Anthropic. Amazon’s actual net income was down, and operating income was flat. Dr. Gaoqing Zhang, an accounting professor at Carnegie Mellon, said that these figures raised “reasonable questions about the subjectivity of these fair-value estimates, given the absence of a public market for Anthropic’s shares.” Dr. Guang Ma, an accounting professor at Rutgers, said that “what Amazon is doing is perfectly fine” from a GAAP perspective. But since fair-value calculations deal with valuation rather than profits—and since Anthropic’s valuation may be inflated—including this gain as a line item was potentially “a loophole for Amazon to exploit.” He added that such decisions historically come down from the C.F.O of a given organization. Similarly, Google disclosed “net gains on equity securities of $10.7 billion,” primarily related to unrealized gains associated with investments in private companies—one of which, according to Bloomberg, was Anthropic. To date, Google has invested some $3 billion in Anthropic, and recently announced a deal to provide the company with a million A.I. chips starting next year, as part of a plan to bring a gigawatt of compute capacity online for Anthropic sometime soon. The deal is reportedly worth tens of billions of dollars. Microsoft’s net income, meanwhile, took a $3.1 billion hit over losses associated with its investment in OpenAI, in which it currently holds a 27 percent stake. (Based on some back-of-the-envelope math, this disclosure means OpenAI lost roughly $11.5 billion this quarter alone.) At the same time, Microsoft revealed that it had delivered $11.6 billion of its $13 billion funding commitment to OpenAI. Last week, the company also said that OpenAI had been contracted to incrementally purchase $250 billion of Microsoft’s Azure services. (OpenAI did not respond to a request for comment.) Microsoft declined to provide details on its “unearned revenue” category, which has swelled to $61 billion. It also wouldn’t address how it accounts for cloud computing credits—which make up the bulk of its investment in OpenAI—or how much of its A.I. computing demand is coming just from OpenAI. “A big part of Microsoft Azure’s growth is driven by OpenAI, but that’s not true for Amazon and Google,” Luria said. “If OpenAI went away tomorrow, Azure wouldn’t be growing 39 percent. Maybe they’d grow 25 percent or 30 percent, but they’ll be fine.” Of course, the question of how the hyperscalers mark the value of their investments in their books—and the proliferation of circular accounting tactics, including cloud computing credits—has become ubiquitous. Amazon, for example, isn’t providing credits to Anthropic as part of its $8 billion investment. But its 2023 deal does stipulate that Anthropic use Amazon Web Services as its primary compute provider. Similarly, in September, Nvidia struck a deal to invest as much as $100 billion in OpenAI, which will in turn buy tens of billions of dollars’ worth of Nvidia chips. Maybe that’s a simple win-win, if everything works out. But some critics worry that the flurry of circular A.I. deals could spell trouble if some of these bets don’t pan out. “I refer to it as unhealthy behavior,” Luria said. “We’re at a point where this is now the common practice: Microsoft, Amazon, Meta, Oracle, Nvidia are all funding usage. It’s legal. It’s been done before. It’s never been done at this scale.”

But one thing that keeps Luria and other economic researchers up at night is leverage. After all, the key difference between the current A.I. investment cycle and past bubbles is that the hyperscalers have been able to divert some of their cash toward big bets on A.I.—but that funding mix is starting to change. To wit, Meta, which generates billions of dollars in free cashflow, recently entered into a lease agreement with Blue Owl Capital to fund a new $27 billion data center; Oracle is in talks for a record $38 billion debt sale tied to data centers; CoreWeave, as Luria noted, is almost “entirely debt financed”; and Nvidia has reportedly had talks about guaranteeing at least some of OpenAI’s data-center-related loans. This is likely just the beginning. “We’re tens of billions of dollars into this debt cycle,” Luria said. “But if OpenAI gets their way, and CoreWeave and Oracle get their way, we’re going to, by the end of next year, be more than $100 billion, maybe even $200 billion into this.”

Again, it might all work out. But if the hyperscalers overbuild, Luria worries, cloud companies may have to reduce the amount they can charge to rent out a G.P.U., which could make it even more difficult for those companies to pay off their debt. “And then you go bankrupt, and then all the debt holders get wiped out,” he said. “And certainly the equity holders of those entities get wiped out.” In many ways, the big four public A.I. hyperscalers—Microsoft, Amazon, Google, and Meta—are too big, too rich, and too diversified to fail. But a significant tech correction would no doubt cascade through other markets, especially at a time when the S&P 500, as my colleague Bill Cohan wrote this week, is being propped up by A.I. “There’s four players at the core of this that will be fine either way, and then everybody else is a marginal player,” Luria said. “I refer to it as bubblicious. We haven’t blown a bubble. We’re in the process of blowing a bubble.”

That’s all for today. I’ll see you on Thursday.

Ian

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St., New York, NY 10006