Now on to the main event…

Jim Chanos, the long-admired short-seller, was on a boat in the Mediterranean when I called him earlier this week, shortly after Meta delivered another gangbusters earnings report. Other investors were practically euphoric: Meta’s quarterly revenue had grown 22 percent year over year, to $47.52 billion, fueled by surprisingly robust advertising income. The stock is now up some 30 percent so far in 2025, and Meta’s market capitalization is closing in on $2 trillion, boosting Mark Zuckerberg’s net worth to more than $260 billion. Pretty much everyone, it seems, was hailing Zuckerberg as a genius for his timely pivot toward A.I.

Chanos, as I have written, currently spends a bunch of his time shorting MicroStrategy, which he believes is overvalued, while simultaneously being long Bitcoin, its principal asset. But he’s also become interested in Meta, although he has no economic stake in the company. And even though he was enjoying some R&R on the high seas, Chanos was happy to talk about his evolving views on the company, now the sixth largest in the world by market capitalization.

In particular, Chanos told me, he has a significant accounting concern. He questioned whether Meta’s depreciation schedule for its massive program of capital spending—primarily on Nvidia A.I. chips—was overstating the company’s net income, and thus, arguendo, inflating its market valuation. In other words, Chanos posited, Meta’s stock could be overdue for a serious correction.

Not to get too technical here—although you guys can handle it—but the idea behind depreciation is to indicate in corporate financial statements how long an asset will have a useful life to the company. For instance, say a company buys some big new piece of factory equipment for $25 million. It’s not unreasonable—and reasonableness is a standard here—to designate that equipment as having a 10-year shelf life. Therefore, you’d capitalize the $25 million purchase on the balance sheet, and then depreciate the value of the asset over its lifetime, taking an annual $2.5 million expense through the income statement for the next 10 years. But if an asset will lose its usefulness in two to three years, as could be the case with Nvidia A.I. chips, depreciating those assets over a longer period of time would mean that expenses on the income statement may be too low—and profits, therefore, too high.

Last week, Chanos observed on X that Meta’s depreciable life on its capital base was 11-12 years, as of the second quarter. But “if the true economic life on its G.P.U.s”—those Nvidia chips—“is actually 2-3 years, most of its ‘profits’ are materially overstated.” (This is not investment advice, but materially overstated is not a good look.) Chanos went on to note that Meta increased its capital base from $145 billion in the second quarter of 2024 to $210 billion in the second quarter of 2025, but the annualized increase in depreciation and amortization was “only $2.8 billion.” His annualized calculation was based on the $700 million increase in depreciation from the second quarter of 2024 ($3.6 billion) to the second quarter of 2025 ($4.3 billion). “That’s 22-year-life, on the margin,” he wrote. “Whatever you think of the stock price here—that’s patently absurd.”

The immense scale of the tech industry’s investments in A.I. can be hard to contextualize. As my new Puck partner Ian Krietzberg noted yesterday, all the major hyperscalers—including Meta, Google, Amazon, and Microsoft—are now devoting absolutely mind-boggling amounts of resources to A.I., with as much as $360 billion in total expenditures expected this year. Building and running data centers isn’t cheap, after all, nor are the tens of thousands of GPUs each one requires. Nvidia’s new Blackwell chips, for instance, are expected to cost between $30,000 and $40,000 apiece.

When Chanos and I connected, via his boat’s Starlink, he pointed me to Meta’s first-quarter filings with the S.E.C.—the second-quarter report hasn’t been filed yet—which suggests that A.I. chips accounted for 85 percent of Meta’s new spending on “property, plant, and equipment,” as indicated by a $7.5 billion increase in “servers and network assets” and a $5.6 billion increase in “construction in progress” costs—“mostly related,” per Meta, to data centers and related network infrastructure. “Accounting is not something that tech investors spend a lot of time on, for generally good reasons,” Chanos told me. “But the numbers now are getting so large that I just think investors need to kind of understand that if the hyperscalers are depreciating their assets over extended periods of time, relative to the economic life of those assets, then by definition they’re overstating their earnings.”

Nvidia C.E.O. Jensen Huang has been boasting that the company’s new A.I. chips are now coming out on a one-year cycle, Chanos noted, rendering older G.P.U.s relatively obsolete, relatively quickly—a trend evidenced by the falling rates at which third parties are renting out their older Nvidia chips, which Chanos is tracking closely. “It’s plummeting,” he said, indicating to him, anyway, that the A.I. chips need to be written off over two to three years, not five to six years. “The whole thing wasn’t material two years ago,” he continued, “but it will certainly be material two years from now, and probably is material as we speak. If Meta was depreciating their assets over Jensen Huang’s chip life, they would have no earnings. And now it’s not quite that bad. But that’s the problem.”

The good folks at Meta, of course, think Chanos is crazy. “I have some issues with what Jim Chanos is saying, and some factual inaccuracies,” Ashley Zandy, a senior director in corporate communications at Meta, told me in an emailed. She urged me not to rely on third-party observations about Meta’s financial judgments and instead pointed me to a footnote in the first-quarter 10-Q, and comments made during the first-quarter earnings call—neither of which, alas, addressed the concerns that Chanos raised about the pace of Meta’s depreciation schedule and how it might be impacting company earnings. When I asked if Meta would like to respond to Chanos’s specific criticisms, Zandy demurred. “We will decline to comment beyond our public filings,” she wrote.

Chanos, ever the contrarian, insisted that investors should be attributing a lower multiple to Meta’s earnings. “I don’t know if it’s going to be resolved by massive writedowns, or what the case might be,” he said, “but one of the ways we can see it actually affecting the business in real time is that free cashflow margins, and actual free cashflows, are beginning to deteriorate. The other sort of interesting observation, as an old guy here, is that these were great software businesses that just gushed huge amounts of cash onto their financial statements—and now they’re becoming the most capital-intensive businesses in the world. They look like the old frackers or auto companies.”

He likened the boom in A.I. chips, and the valuations of the companies that make them and use them, to the boom that occurred a generation ago in the infrastructure companies that built out the internet, and what happened when the spending spigots shut off. He said it’s important to remember that while, say, Meta, is expensing these chips relatively slowly, as compared to their useful life, Nvidia is taking the sale of these chips into its revenue and profits all at once. “In terms of the cycle itself, you get a mismatch of dollars, disproportionately benefiting profits to a certain small number of companies,” Chanos said. “In 1999 and 2000, it was the Ciscos and Nortels and Lucents. Today it’s Nvidia. So [Nvidia] gets to book revenues and profits. Meta and Microsoft and Amazon get to capitalize a lot of that spending.”

Chanos wouldn’t offer any predictions about how this crazy cycle will end, but he’s seen these movies before. And as a short seller, he figures he knows a top when he sees one. “Nobody’s worried about this stuff now,” he told me. “But they should be.”

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