It was probably inevitable that Michael Burry, the ur-contrarian of modern finance, patron saint of the uncomfortable truth, would eventually take aim at the great secular religion of our time: artificial intelligence.
Having made his reputation by spotting the rot beneath America’s subprime housing boom, Burry has now trained his eye on another corner of exuberant finance, and he does not like what he sees.
In recent weeks the Scion Asset Management founder has disclosed sizeable short positions against Nvidia and Palantir, two of the market’s most zealously worshipped AI totems. And now he has fired a fresh salvo, accusing the hyperscalers, the Silicon Valley giants building the world’s AI plumbing, of playing fast and loose with their accounts. In particular, he claims they are using aggressive depreciation assumptions to flatter earnings, an old trick wearing new clothes.
“Understating depreciation by extending useful life of assets artificially boosts earnings — one of the more common frauds of the modern era,” Burry posted on X, noting that companies buying billions in chips with a two-to-three-year product cycle have magically decided these machines will last five or six. “Yet this is exactly what all the hyperscalers have done.”
His point is simple enough: when Nvidia’s NASDAQ:NVDA bleeding-edge GPUs are obsolete almost before the shrink-wrap hits the floor (replaced first by Blackwell, soon by Rubin and Rubin Ultra, each several multiples more powerful) the idea that these machines will hum along profitably for half a decade is, at best, optimistic. At worst, it is the sort of optimism that ends abruptly in the small print of a restated earnings report.
Burry’s own maths suggests that these extended depreciation schedules will inflate industry profits by a cool $176 billion through 2028. He singles out Oracle NYSE:ORCL and Meta as the more egregious beneficiaries: by 2028, he reckons, Oracle’s earnings will be overstated by 26.9%, Meta’s by 20.8%. Investors, he suggests, are not just buying into the AI dream, they may be buying into the accounting.
- The AI concentration risk nobody wants to discuss
- Artificial Intelligence: boom, bust or just beginning?
- Seeing Machines lands breakthrough North American deal
Markets are starting to twitch about AI valuations
His scepticism comes as markets begin to twitch. Earlier this month CoreWeave, the once-darling AI cloud provider, was hammered after disappointing guidance. The company, Burry notes, also extended depreciation cycles on its Nvidia chips, many of them Hoppers, state-of-the-art only last year but now yesterday’s technology. In a world where utilisation rates of 60–70% eat through hardware like moths through wool, the idea of six-year GPU lifespans borders on the fantastical.
For the AI bulls, depreciation is a footnote in a saga of limitless technological promise. For Burry, it is the heart of the story. High-flying valuations rest on the assumption that today’s monumental capex binge will eventually throw off profits. If those profits are an artefact of accounting rather than economics, the downside could be rather greater than the optimists admit.
Of course, Burry being Burry, he is not merely muttering from the sidelines. His disclosed short positions amount to more than $1.1 billion notional, with Palantir, the quasi-mystical darling of the AI defence-tech crowd, the larger bet. These aren’t hedges. They’re statements.
Is the AI boom a bubble then?
To be clear, his Nvidia and Palantir shorts are ‘separate’ from his depreciation critique; the former are tactical trades, the latter a systemic challenge. Together, however, they form a clear message: the AI boom is starting to smell like a bubble. The analogies to the dotcom era come quickly. Burry himself highlights a 2002 note observing that less than 5% of America’s telecoms fibre capacity was in use after the bust, an object lesson in what happens when capex gets ahead of reality.
Critics will say he’s early. They may well be right. Momentum is a stubborn force, and bubbles burst only after they have exhausted every last pocket of disbelief. But Burry’s warning is characteristically blunt: behind the optimism lies an earnings structure built on heroic assumptions about hardware that ages like milk.
On 25 November, he promises more detail. Whether markets will still be listening, or whether the AI faithful will dismiss him as a ghost of bubbles past, remains to be seen. But if history is any guide, it pays to read the footnotes before the headlines.
Comments (0)