As the AI arms race intensifies, a trend of borrowing to expand by using NVIDIA GPUs as collateral is gaining momentum on Wall Street. However, a renowned short-seller who accurately predicted the Enron collapse has issued a stern warning regarding this practice…
Emerging AI cloud computing companies, in order to challenge tech giants in the artificial intelligence market, have secured substantial loans collateralized by graphics processing units (GPUs) from a chip manufacturer valued at $5 trillion. These funds are used to purchase additional AI chips to scale operations. These companies lease a data center, fill it with AI hardware, and rent out the computing power of this hardware to tech companies for training and running artificial intelligence models.
According to a report by The Information in July, four companies in this growth sector — three of which have $NVIDIA (NVDA.US)$ investment backgrounds, namely leveraging NVIDIA’s AI chips (i.e., GPUs) as collateral — have accumulated over $20 billion in debt.
Jim Chanos, renowned for accurately predicting the Enron collapse, sees warning signs in this surge of GPU-backed debt, as these “neoclouds” lack a clear path to profitability, making debt repayment challenging.
“Business models like those of neoclouds, and many AI companies themselves…are currently just loss-making enterprises,” Chanos said in an interview last week. “You can only hope that this changes; otherwise, these debts will face default.”
This financing trend, pioneered by CoreWeave, represents a form of asset-backed lending: loss-making companies can use their assets as collateral to obtain funding they would otherwise be unable to access, typically at higher interest rates.
As so-called neoclouds are much smaller in scale compared to the tech giants they aim to challenge, GPU-backed financing provides an effective pathway for rapid expansion. NVIDIA has heavily invested in this area, aiming to expand its customer base beyond tech giants.
According to The Information, CoreWeave, which went public earlier this year and is backed by NVIDIA,$CoreWeave (CRWV.US)$ , and its competitor cloud service provider Fluidstack each have approximately $10 billion in debt collateralized by their GPU reserves. Similarly, NVIDIA-backed Lambda and Crusoe carry $500 million and $425 million, respectively, in GPU-collateralized debt.
CoreWeave reported a loss of about $65 million in 2024, while Fluidstack’s losses for the same year were less than $1 million. Financial data for Lambda and Crusoe remain undisclosed.
Another significant risk associated with AI chip-backed debt is a topic of intense debate within the technology investment community: NVIDIA’s GPUs could depreciate, or lose value faster than accounted for in company financials.
Most cloud providers, including CoreWeave, expect their AI chips to generate revenue over approximately six years. In January, Amazon reduced the depreciation period — or effective economic life — of its AI chips from six years to five.
However, Chanos noted that with NVIDIA launching new AI products roughly every 18 months, the rate of chip depreciation could accelerate. First, this would make it more difficult for companies to earn sufficient revenue from the chips to repay their debts. Second, the market value of the chips could fall below the value of secured loans, thereby increasing the risk of default.
Chanos pointed out that even assuming a long lifespan for the chips, new cloud providers are currently unable to turn a profit: “Many of the business models we observe, including some new data centers and emerging cloud providers like CoreWeave and NEBIUS, are not profitable even when depreciating assets over eight or ten years,” he said. “The cost of capital remains too high, surpassing the return on AI data center transactions.”
“If the economic life of these (NVIDIA's chips) is only three years… then the overall economic logic of many such deals will completely collapse,” Chanos said.
Editor/Liam