Michael Burry has opened short positions against Oracle and Nebius Group, arguing both companies understate how fast their AI chips lose value and therefore overstate earnings. He estimates Oracle's earnings could be overstated by approximately 26-27% by 2028 and has bought put options on the stock. Nebius, meanwhile, holds AI compute contracts worth up to $27 billion from Meta.
Michael Burry, best known for calling the 2008 housing crash in "The Big Short," is betting against two AI infrastructure companies: Oracle and Nebius Group. His argument is that neither company depreciates its GPUs fast enough, so their reported earnings look stronger than the hardware's real economics support.
The comparison he draws is blunt. Underestimating depreciation works like assuming a car holds its value for ten years — the fiction inflates net worth on paper while the underlying economics haven't changed. Burry applies that logic to billions of dollars in AI chips sitting on hyperscaler balance sheets.
Oracle's depreciation gap
Burry estimates Oracle's earnings could be overstated by approximately 26-27% by 2028 because of undervalued GPU depreciation. He has taken put options on Oracle shares after shorting the stock in recent months, a leveraged bet that the shares will fall further. He has pointed to the industry's projection of $176 billion in AI-related capital expenditures across hyperscalers as context for how large the gap could grow.
Nebius's AI compute contracts
Nebius Group is a less household name than Oracle, but it has become a significant player in AI compute. The company has secured AI compute contracts totaling up to $27 billion with Meta over five years, and it also works with Microsoft. Burry's skepticism toward Nebius appears rooted in thematic concern about the AI infrastructure sector broadly; no reliable regulatory filings confirm a new short position against Nebius beyond that thematic bearishness.
A closed fund, an open bet
Burry's hedge fund, Scion Asset Management, was liquidated by late 2025. He has kept sharing his market analysis through a Substack publication, maintaining a bearish outlook on AI-related investments.
Technology companies retain wide discretion in choosing useful-life estimates for their hardware, and a GPU depreciated over five years produces very different financial statements than the same GPU depreciated over three years, even though the underlying economics haven't changed.
Source: Crypto Briefing
Trading involves risk.
Risk Disclaimer
