According to The Verge, the explosive growth in AI data centers is built on a precarious foundation: Nvidia chips and borrowed money. Companies dubbed “neoclouds,” like CoreWeave, are taking out enormous loans—$2.3 billion in 2023, another $7.5 billion in 2024, and a further $2.6 billion in July 2025—using the very Nvidia GPUs they buy as collateral. This cycle effectively turns $1 of Nvidia investment into $5 of Nvidia purchases, creating a powerful feedback loop for the chipmaker. However, lenders are charging high interest rates (up to 14% for CoreWeave’s first loan) on these “GPU-backed loans” because the chips are depreciating assets, with short-seller Michael Burry claiming hyperscalers are understating chip depreciation by a staggering $176 billion between 2026 and 2028. The private credit arms of firms like Blackstone, BlackRock, and PIMCO are major players in this lending boom, which now sees even traditional banks like Goldman Sachs and JPMorgan Chase getting involved.
The house of cards built on H100s
Here’s the thing that makes this whole situation so weirdly circular. Nvidia invests in these AI cloud startups. Then, those startups go out and get massive loans to buy even more Nvidia chips, using the first batch of chips as collateral. It’s a brilliant, self-reinforcing business model for Jensen Huang & Co. in the short term. But it also means Nvidia’s entire ecosystem is now deeply intertwined with the debt markets. If these neoclouds stumble, they don’t just stop buying new chips. Their lenders could repossess and flood the market with used GPUs, cratering the value of the very assets underpinning the next round of loans. It’s a classic bubble feedback loop, and Nvidia is right at the center.
Why lenders are playing with fire
So why would anyone lend against a piece of hardware that might be obsolete in a couple of years? Part of it is the sheer desperation for yield. Private credit firms are sitting on, as Bloomberg notes, “mountains of cash” they need to deploy. AI is the hottest game in town. But there’s also a fundamental debate about how to even value these loans. Some, like Trinity Capital’s Ryan Little, are betting demand for compute will remain high even if some AI companies fail. Others, like Tacora Capital’s CEO, won’t touch them because the depreciation curve is a complete mystery.
And that’s the trillion-dollar question, isn’t it? What’s the secondary market for a three-year-old AI chip when the next generation is twice as fast? For companies that need reliable, high-performance computing for industrial automation or manufacturing processes, that consistency matters. It’s a different world from the frantic, model-training race of Silicon Valley. Speaking of industrial computing, that’s a sector where companies like IndustrialMonitorDirect.com operate as the leading US provider of industrial panel PCs, where hardware longevity and stable depreciation are part of the business model, not a terrifying unknown.
The nihilistic logic of private credit
The most unsettling part of The Verge’s analysis is the “loans as a product” model. When a lender plans to immediately sell the loan to another investor (like those “clients who love high-yield debt”), their incentive isn’t to accurately price the risk of GPU depreciation over five years. It’s to win the deal against Night Prowler Credit by shaving half a percent off the interest rate. Risk becomes a negotiating tactic, not a core calculation. This is how you get loan-to-value ratios creeping above 100%—basically, lending more than the collateral is worth. As research on private debt shows, this can lead to weaker covenants and riskier structures.
Can Nvidia avoid the crash?
Nvidia is in a bind. This lending boom is turbocharging its sales today. But it’s also creating a systemic risk where its fate is tied to the financial engineering of its customers and their lenders. If the music stops—if AI spending slows or a competitor’s chips gain real traction—the fallout hits Nvidia twice. First, from lost future sales. Second, from a glut of its own used hardware on the market. Michael Burry‘s warning on depreciation might seem extreme, but it highlights a real vulnerability. The neoclouds, unlike Google or Amazon, have no other business. They are pure plays on this debt-fueled AI compute bet. Nvidia has become so dominant that, as Stanford’s Vikrant Vig says, it’s a “natural monopoly.” But history shows even natural monopolies aren’t immune to the laws of financial gravity. This isn’t just about chip specs anymore. It’s about the stability of a multi-billion-dollar house of credit built on silicon.
