Michael Burry, the investor who famously predicted the 2008 financial crisis and shorted the US housing market before it collapsed, has turned his attention to Big Tech’s AI spending binge—and he’s asking a question nobody on Wall Street seems to want to answer: when does it actually stop?In a series of posts on X over the weekend, Burry—who runs Scion Asset Management and goes by the handle Cassandra Unchained—called out Amazon, Google, Meta, Microsoft, Oracle, Nvidia, and Caterpillar by name. His core argument is blunt. These companies are burning through all their cash flow on AI data centres, borrowing money in ways they never have before, and using accounting tricks to mask the damage to their earnings. “It is consuming all your cash flow, you are borrowing, you are financing in ways you never have,” Burry wrote. “But if it scales, when does it end?”
Big Tech’s combined AI tab has crossed $660 billion—and the debt is piling up
The numbers back him up. Amazon, Alphabet, Meta, and Microsoft are collectively set to spend roughly $660 billion on AI infrastructure in 2026—a figure that’s 165% higher than what they spent in 2024 and larger, as a share of US GDP, than the Apollo space programme, the interstate highway build-out, and the great railroad expansion of the 1850s. Amazon alone pledged $200 billion, a number so large it’s expected to exceed its own cash from operations for the first time. The company has already filed paperwork signalling it may raise fresh capital through debt.Alphabet is issuing a 100-year century bond—the first by a tech company since IBM in 1996—as part of a $20 billion-plus debt raise across multiple currencies. Its long-term debt has quadrupled to $46.5 billion. Meanwhile, the Stargate AI data centre project—announced with a $500 billion target involving OpenAI, SoftBank, and Oracle—is reportedly stalled, with under $10 billion in secured funding and no major construction underway.Wall Street has not taken this well. Amazon, Google, and Microsoft have shed a combined $900 billion in market value since their earnings calls. Microsoft disclosed that 45% of its $625 billion cloud backlog comes from a single customer—OpenAI—which unnerved analysts. And as Burry pointed out, the combined revenues of Amazon, Apple, Alphabet, Microsoft, Meta, and Nvidia don’t even hit $2 trillion. The gap between what’s coming in and what’s going out explains why leverage is the new normal.
Burry draws a parallel to the early electricity boom—and that’s not entirely reassuring
When pushed on whether the AI buildout is just another Cisco-style bubble, Burry offered a more nuanced comparison: electrification in the late 19th and early 20th century. The technology was real, demand was enormous, and the killer app—electric light—was obvious. But as Burry’s own research showed, citing the academic paper “Killing Complaints with Courtesy,” even iconic names like Westinghouse and Edison went through bankruptcies, price wars, and brutal write-downs before the industry stabilised.Countless central power stations went bankrupt within their first four decades. The panics of 1893 and 1907 wiped out many companies, and even Westinghouse Electric itself filed for bankruptcy. Edison, for his part, eventually lost interest entirely, saying he simply wanted dividends from whatever stock he still held. Electricity production surged 280% between 1900 and 1910, and generation capacity grew 375%—but that didn’t save the investors who funded it too early or too aggressively.Google CEO Sundar Pichai, speaking at the India AI Impact Summit 2026 in Delhi, reached for a similar historical comparison—railroad expansion and the US highway system—to justify Alphabet’s spending. But it’s worth noting that just two months earlier, in November, Pichai himself had told the BBC there were “elements of irrationality” in the AI investment boom and drew a direct parallel to the dot-com crash. The pivot from caution to conviction has been swift.
The depreciation trap: today’s cutting-edge GPU is tomorrow’s write-off
Then there’s the hardware problem. Nvidia’s own roadmap shows its upcoming Vera Rubin chips delivering roughly 5x the inference performance of current Blackwell GPUs—in just one generational jump. That kind of leap makes a six-year depreciation schedule look like wishful thinking. Microsoft CEO Satya Nadella has himself said he doesn’t want to be stuck depreciating one generation of hardware for four to five years.Burry’s analysis estimates that if chip useful life is closer to 2.5 years, cumulative earnings overstatement across the Big Tech cohort could average 32% between 2026 and 2028. Even at 3 years, the average overstatement sits at 18%. Oracle comes out worst in his model, with potential overstatement of 62% at a 2.5-year useful life. These aren’t small rounding errors—they’re the kind of gaps that eventually force large write-downs. And a $50 billion write-off, as Burry noted, might not sound like much against a $3 trillion market cap today—but it will loom a lot larger if that market cap shrinks.Burry isn’t saying AI is fake or that the technology won’t matter. His electrification analogy makes that clear enough. What he’s saying is that the financial engineering around the buildout—the debt, the depreciation schedules, the earnings adjustments—deserves far more scrutiny than it’s getting. The technology was transformational then too. It just didn’t stop the investors from getting wiped out.
