Nvidia-backed loans to AI data center operators, including CoreWeave, have raised concerns about financial stability as the sector relies on depreciating graphics chips as collateral.

The AI data center build-out depends on Nvidia chips and large amounts of debt. Companies such as CoreWeave have used the chips as loan collateral, securing billions of dollars from private credit firms.

CoreWeave obtained a $2.3 billion loan in 2023, followed by $7.5 billion and $2.6 billion extensions in 2024, with interest rates around 14 percent and loan‑to‑value ratios ranging from 50 percent to more than 100 percent, according to Trinity Capital and company filings.

Vikrant Vig, a finance professor at Stanford, said that Nvidia’s chips are more liquid than other GPUs, which can lower financing costs, but the reliance on the chips as collateral creates a natural monopoly for Nvidia.

Analysts question how long the chips retain value. Michael Burry has argued that hyperscalers understate GPU depreciation by $176 billion between 2026 and 2028, while companies such as CoreWeave state that loan capacity depends on the chips’ worth, which may decline faster than projected.

Keri Findley, CEO of Tacora Capital, said she does not evaluate GPU‑backed loans because she views depreciation as unclear, and many neoclouds may have to increase collateral or face write‑downs if chip values fall.

Private credit firms, which are less regulated than banks, have deployed billions of dollars into AI‑related debt, creating exposure that could spread to banks if multiple neoclouds default; Moody’s chief economist Mark Zandi warned that such debt could threaten the broader financial system.

Nvidia has a strong incentive to keep neoclouds afloat, as its revenue depends on sales to large customers and its own cloud contracts, and it has previously supported CoreWeave’s IPO.

The sustainability of the GPU‑backed loan model will depend on how quickly chip values depreciate and whether AI spending continues to grow, an outcome that remains uncertain.