Prominent short-seller Jim Chanos raised alarms this week about a growing corner of the debt market where AI cloud companies have borrowed more than $20 billion using Nvidia graphics processing units as collateral, warning that rapid chip obsolescence could trigger a wave of defaults.
In an interview with Yahoo Finance published December 1, Chanos—known for his prescient call on Enron’s collapse—said “neocloud” companies operating AI data centers are “currently just loss-making entities” that may struggle to repay loans if their GPU assets depreciate faster than expected. “There’s going to be debt defaults,” he stated.
Debt Concentrated Among Four Companies
CoreWeave and Fluidstack each carry roughly $10 billion in GPU-backed debt, while Lambda and Crusoe hold $500 million and $425 million respectively, according to reporting from The Information cited across multiple outlets. These firms, several backed by Nvidia investments, lease space in data centers, install AI chips, and rent computing power to tech companies training AI models.
The financing model represents a type of asset-based lending that allows unprofitable companies to secure funding using their GPU reserves as collateral, often at elevated interest rates. CoreWeave , which went public in March at $40 per share, reported a loss of approximately $65 million in 2024 while carrying around $14 billion in total debt as of the third quarter.
Depreciation Concerns Mount
Chanos highlighted a critical risk: Nvidia releases new AI chip architectures approximately every 12 to 18 months, potentially rendering older models obsolete before companies can generate sufficient revenue to cover their debt obligations. Most cloud enterprises, including CoreWeave , estimate their AI chips will produce revenue for about six years, though Amazon recently reduced its depreciation timeline from six to five years.
“If the economic lifespan of these Nvidia chips is only three years, the financial viability of many of these agreements collapses,” Chanos said.
The warnings come amid broader scrutiny of Nvidia’s ecosystem. Investor Michael Burry recently compared the chipmaker to Cisco Systems during the dot-com bubble, arguing that AI infrastructure is being overbuilt before demand materializes. Nvidia has pushed back against such comparisons, calling them misrepresentations and emphasizing genuine demand from large cloud providers.