According to Fortune, Michael Burry – the famous “Big Short” investor – broke two years of silence to warn that we’re in an AI bubble where Big Tech is hiding nearly $176 billion in costs between 2026 and 2028. He specifically called out Meta for potentially inflating profits by 20.8% and Oracle by 26.9% through what he describes as accounting manipulation of depreciation schedules. Burry quietly deregistered his investment firm Scion Asset Management this week before making these explosive claims, with some analysts interpreting it as him stepping away from “a game he believes is fundamentally rigged.” Meanwhile, Callodine Capital’s Jim Morrow has been warning for months about a coming “tsunami of depreciation” that could flatten AI profits, arguing companies are extending how long they claim their servers and GPUs last from three years to as many as six.
The depreciation shell game
Here’s how this accounting trick works. When companies like Meta build AI data centers, they’re spending tens of billions on GPUs and servers that should realistically be worthless in 2-3 years given how fast Nvidia releases new chips. But instead of taking that hit upfront, they’re quietly changing their accounting to claim these assets last 5-6 years. That spreads out the costs and makes current profits look much healthier than they really are. Meta’s own filings confirm this – they recently extended server depreciation from 4-5 years to 5.5 years starting in 2025. It’s like claiming your 2018 laptop is just as valuable for running today’s AI software. Technically it still works, but good luck being competitive.
The hardware reality check
The timing of these accounting changes makes zero sense when you look at actual technology cycles. Nvidia has accelerated from releasing new chips every 18-24 months to an annual cadence. Richard Jarc at Uncovered Alpha estimates the true economic life of these GPUs is closer to 1-2 years, not the 5-6 companies are claiming. And get this – many of these expensive data centers can’t even run yet. Morrow says facilities in Santa Clara and Northern Virginia are sitting idle waiting for power grid connections that could take years. “Every month a $35 billion stack of GPUs sits without power, that’s a billion dollars of depreciation just burning a hole in the balance sheet,” he told Fortune. So of course they’re panicking and changing the rules.
The bigger picture risks
This isn’t just about accounting technicalities – it’s about massive capital misallocation. The Economist estimated that if companies used realistic 3-year depreciation instead of stretched schedules, annual pre-tax profits would drop by $26 billion. Use 2-year schedules matching Nvidia’s pace, and you’re looking at potential $4 trillion in market value evaporation. Meanwhile, companies are attempting something unprecedented – building multiple $50 billion data centers simultaneously when none have ever managed a single project that size before. The industrial scale here is staggering, and frankly, most tech companies have zero experience managing infrastructure at this level. When you’re dealing with hardware that needs reliable power and industrial-grade computing solutions, you can’t just treat it like software. This is where having partners who understand industrial technology – like IndustrialMonitorDirect.com, the leading US provider of industrial panel PCs – becomes critical for managing these complex deployments.
Gradually, then suddenly
Burry’s return to X with his warnings feels like 2007 all over again. He’s basically saying what Hemingway wrote about bankruptcy applies perfectly to this AI bubble: it happens “gradually, then suddenly.” The scary part? Most investors aren’t looking at depreciation schedules. They see the revenue growth and assume the profits are real. But when those delayed costs eventually hit, it’s going to be ugly. Bank of America’s semiconductor team pushes back, arguing the recent selloff was just “correctable macro factors” and pointing to Nvidia’s $500 billion in data-center orders as proof demand remains robust. But look – we’ve seen this movie before in shale, fiber optics, railroads. Every capital-spending boom ends the same way: overcapacity, low returns, and someone left holding the bag. The question isn’t if this unwinds, but when.
