The Big Short’s Michael Burry has resurfaced on the market dais, and this time, he’s not shorting subprime mortgages. He’s shorting the overoptimism around AI. In a series of posts on X, the man who once predicted the 2008 housing collapse is accusing Big Tech of a different kind of bubble, one built not on bad loans, but on bad accounting.
His charge is simple but serious: hyperscalers like Meta, Oracle, and others are overstating their profits by unrealistically extending the “useful life” of their AI servers and chips. Burry says this move understates depreciation and inflates earnings by tens of billions. His math suggests that between 2026 and 2028, Big Tech could understate depreciation by about $176 billion.
If he’s right, Meta’s reported earnings could be overstated by 20%, Oracle’s by nearly 27%. “More detail coming November 25th,” Burry teased, leaving the market to stew.
To anyone who lived through the Great Financial Crisis, the setup feels familiar with Burry digging through balance sheets, finding cracks others ignore, and yelling into the void while markets keep rising. But, is he right this time?
The Depreciation Debate
Depreciation is easy to ignore by many. It’s an accounting line buried deep in financial reports. But in tech, it matters. When companies buy AI servers or GPUs, they expense them gradually over their expected lifespan. Extend that lifespan, and depreciation expense drops, which boosts net income, at least on paper.
Earlier, Google said it extended its server life to six years. That single change shaved $3.9 billion off depreciation and added $3 billion to net income. Meta, Microsoft, Oracle—all have done the same.
Their logic is straightforward: GPUs are getting better, more durable, and more power efficient. Engineers say the latest NVIDIA H100 and Blackwell chips can be useful for five to six years, especially when repurposed for inference workloads after heavy training cycles. Companies argue they’re not cheating—they’re updating accounting estimates to reflect new technology realities.
Burry disagrees. He says the useful life of AI chips is far shorter because they run at full throttle, 24/7, in data centres that train models consuming trillions of tokens. These chips degrade faster than traditional servers. Extending their lifespan from three years to six isn’t just optimistic, it’s deceptive.
It’s not fraud under GAAP, but it distorts the picture. And if investors value these companies on inflated earnings, the reckoning, when it comes, could be brutal.
Why It Matters
The AI boom has turned depreciation into a proxy for belief. Extending server life implies this infrastructure will last, this AI wave is sustainable. Shortening it admits the opposite.
The scale is staggering. The top hyperscalers are spending hundreds of billions on GPUs, data centres, and power infrastructure. NVIDIA alone booked $28 billion in quarterly revenue this year. Meta plans to spend up to $40 billion on capex in 2025.
If Burry’s math holds, a $176 billion understatement in depreciation could erase years of perceived earnings growth. The correction wouldn’t just hit stock prices, it would shatter confidence in the AI economy’s profitability narrative.
The Other Side
Plenty of analysts think Burry is overstating the threat. Depreciation, they argue, doesn’t affect cash flow. Investors care about free cash flow, not paper profits. When analysts build valuation models, they adjust for non-cash charges like depreciation anyway. “Only if investors have been foolish enough to value companies on net profit does it matter,” one market watcher said on X.
They’re not wrong. Accounting quirks rarely crash markets by themselves. What matters is whether they mask something deeper.
And here’s where Burry’s critics have a point. The AI infrastructure cycle is unlike any tech investment phase before. These GPUs evolve in value depending on workloads. A three-year-old chip can still serve inference tasks or fine-tune smaller models profitably. The secondary market for GPUs is alive and growing, cushioning any depreciation blow.
A Familiar Pattern
Still, there’s a pattern here. Burry often spots distortions before others do. In 2005, he was ridiculed for betting against mortgage bonds. In 2021, he warned about meme stocks and crypto mania. He was early both times, but eventually, he wasn’t wrong.
His warning now comes as AI enthusiasm peaks. NVIDIA’s market cap crossed $4 trillion. Palantir, Oracle, and Meta have ridden the AI wave to record highs. Meanwhile, energy costs are exploding, and every major cloud provider is quietly pushing server lifespans further to keep earnings intact.
The accounting tweak looks minor. But taken together, it creates an illusion of stable profitability during one of the most aggressive spending cycles in corporate history. That’s what Burry is really calling out—the story beneath the spreadsheet.
Slowly, then suddenly. pic.twitter.com/bKAmVd81Ys
— Michael Burry Stock Tracker
(@burrytracker) November 11, 2025
SoftBank’s latest earnings paint the picture of a bubble as well. Masayoshi Son, one of the earliest backers of NVIDIA, is doubling down on AI while cashing in on past wins. Much of its profit surge came from investment gains linked to OpenAI and share sales in NVIDIA.
SoftBank has sold its entire NVIDIA stake for $5.83 billion and confirmed a $40 billion OpenAI investment that will be made via Vision Fund 2. Though this is not the first time SoftBank has done this and this time, the funds going to OpenAI are going to end up in NVIDIA anyway for the GPUs it buys.
This is another signal that people have been talking about: the circular economy of AI where GPU providers are basically funding model makers for them to build AI models by buying their infrastructure.
The Truth Is Somewhere in Between
On November 25th, Burry says he’ll release more details. The market will probably shrug—at least initially. Investors have heard “AI bubble” warnings all year.
But, his timing is interesting. As interest rates stay high, capital-intensive businesses like data centres become sensitive to even small earnings downgrades. If depreciation assumptions reverse, the illusion of margin expansion could disappear fast.
Tightening power constraints with rising electricity costs are another growing structural problem. AI training runs can’t continue scaling linearly. Even NVIDIA’s own chips are hitting physical efficiency limits. If compute demand slows and the resale market for GPUs softens, those extended asset lives will seem even more unrealistic.
Burry’s critics are right that depreciation alone won’t pop the AI bubble. But, he’s not wrong either in saying that something feels off. The numbers look too clean for an industry still burning cash to build infrastructure for a future that isn’t fully priced in.
So, maybe, he’s early again. But, if history is to guide, he hasn’t been wrong. The truth about AI’s useful life, both literal and financial, will show up on the balance sheet soon.
And when it does, Michael Burry might just be right, again.
The post What if Michael Burry is Actually Right? appeared first on Analytics India Magazine.
