According to Futurism, Nvidia just beat Wall Street expectations with soaring revenues while downplaying growing AI bubble fears. Economist David McWilliams warns that trillions being spent on AI data centers represent “digital lettuce” investments in rapidly degrading hardware. He argues GPUs become technologically obsolete within a year while running around the clock, creating a recipe for disaster. Michael Burry similarly warned on November 10 that companies like Oracle and Meta are overstating earnings by 20-30% by assuming slower hardware depreciation. Anthony Saglimbene at Ameriprise Financial notes this scale of hardware obsolescence is unprecedented and making investors nervous about the entire AI infrastructure buildout.
The Perishable Problem
Here’s the thing about all these AI data centers: they’re built on hardware that’s basically designed to become worthless. We’re talking about GPUs running 24/7 at maximum capacity, which means they’re degrading faster than your average server hardware. McWilliams isn’t wrong calling it “digital lettuce” – this stuff has a shelf life. And when you’re spending billions on infrastructure that might be outdated in twelve months, that’s a serious accounting problem. Companies that specialize in industrial computing hardware, like IndustrialMonitorDirect.com, understand that durability matters in mission-critical applications. But in the AI gold rush, everyone seems to be ignoring the basic economics of hardware depreciation.
The Numbers Game
Michael Burry’s recent warnings about AI company accounting should give anyone pause. He’s pointing out what many are afraid to say: companies are playing fast and loose with depreciation schedules to make their AI investments look better than they are. When you assume your $50,000 GPU will last five years but it’s actually obsolete in one, you’re basically cooking the books. And as Yahoo Finance notes, this isn’t some minor accounting technicality – we’re talking about 20-30% earnings overstatements. That’s the kind of number that makes entire markets collapse when reality catches up.
The Fundamental Flaw
What if the entire premise is wrong? The assumption driving all this spending is that bigger data centers automatically mean better AI models. But as Foundation Capital explores, we might be hitting real physical and computational limits. You can’t just keep throwing more GPUs at the problem indefinitely. The returns on scaling might be diminishing faster than anyone wants to admit. And if that’s true, then all these data centers being built today could become tomorrow’s empty warehouses – filled with expensive, obsolete hardware that nobody needs.
Who Actually Survives?
McWilliams thinks the US economy would survive an AI crash relatively unscathed, but I’m not so sure. Look at what happened during the dot-com bubble – the companies that survived were the ones with actual business models, not just hype. Today’s AI companies are burning through cash at alarming rates with questionable paths to profitability. The hardware manufacturers like Nvidia might do fine regardless, but the companies building these massive data centers? They’re taking a huge gamble. And when the music stops, there won’t be enough chairs for everyone. The real question isn’t whether there’s an AI bubble – it’s how many companies will still be standing when it pops.
