Legendary short-seller Jim Chanos, founder of Kynikos Associates and the investor who famously predicted Enron’s collapse, is sounding the alarm on a critical segment of the AI investment landscape. In a series of posts on X, Chanos argued that investors should focus on companies building AI models rather than data center operators that power them, warning that the data center boom may not be sustainable.
Chanos’s skepticism comes amid the data center construction frenzy of 2025, as tech giants rushed to build infrastructure supporting their AI ambitions. His comments followed Nvidia’s announcement of a $2 billion investment in CoreWeave, an AI infrastructure company that has seen its stock rally but faces questions about its path to profitability.
The short-seller’s investment thesis is straightforward: “The former are technology companies, the latter are REITs.” He argues that while data center companies may appear to be tech firms, economically they function more like real estate investments. This distinction is crucial for investors trying to capture AI’s growth potential.
Using CoreWeave as a case study, Chanos questioned whether investors are examining actual fundamentals or simply buying into AI hype. He pointed out that based on CoreWeave’s annualized third-quarter results, the company would still be reporting losses even when using a 10-year depreciation schedule for its GPU assets. CoreWeave’s recent earnings report showed revenue exceeding Wall Street estimates, but the company scaled back guidance for the coming year and revealed temporary delays with a data center partner.
Chanos’s concerns extend beyond individual companies to the broader AI infrastructure buildout. He highlighted the massive capital expenditure flowing into physical AI infrastructure—data centers, chips, and related hardware—and warned of potential consequences if companies begin scrutinizing their return on investment. “If anyone decides to pause and ask, ‘What’s our real economic return here?’ it could be a big problem,” he stated in a previous interview.
The investor’s perspective represents a contrarian view in a market where AI infrastructure stocks have been among the hottest investments, suggesting that the economics of data center operations may not justify current valuations.
Key Quotes
In that case you should own the companies that build the models, not the companies that build the data centers. The former are technology companies, the latter are REIT’s.
Jim Chanos articulated his core AI investment thesis in response to a user discussing their willingness to pay for AI coding tools. This statement encapsulates his view that data center operators are fundamentally real estate plays rather than technology investments.
Does anyone still bother to check these wildly bullish claims with, you know, their actual financial statements…?! Because [CoreWeave] based on its annualized 3Q results, would still be reporting losses USING 10-YR LIFE for its GPU depreciation!
Chanos questioned CoreWeave’s financial fundamentals despite bullish market sentiment, highlighting that even with generous depreciation assumptions, the company remains unprofitable. This illustrates his concern that investors are ignoring basic financial analysis in favor of AI hype.
I’m starting to worry there’s so much spending right now on the AI physical boom — the buildout of data centers, chips, and so on — that if anyone decides to pause and ask, ‘What’s our real economic return here?’ it could be a big problem.
In a previous interview, Chanos expressed concern about the sustainability of massive AI infrastructure spending. This warning suggests potential systemic risk if companies begin questioning whether their capital expenditures will generate adequate returns.
Our Take
Chanos’s perspective introduces much-needed scrutiny to the AI investment frenzy. His track record of identifying overvalued assets makes his warnings impossible to ignore. The distinction between AI model builders and infrastructure providers is particularly astute—companies like OpenAI, Anthropic, and Google’s DeepMind create defensible intellectual property and recurring revenue streams, while data center operators face commoditization risk, massive capital requirements, and rapid hardware obsolescence.
The GPU depreciation issue Chanos highlights is especially concerning. With AI chip technology evolving rapidly, today’s cutting-edge hardware may become obsolete within years, not the decade-long depreciation schedules suggest. This creates a potential accounting time bomb for infrastructure companies. As the AI market matures and competition intensifies, data center operators may face pricing pressure while still servicing enormous debt loads from their buildouts. Investors should heed Chanos’s advice and carefully distinguish between companies capturing AI’s value creation versus those simply providing commoditized infrastructure.
Why This Matters
Chanos’s warning carries significant weight for the AI investment landscape, particularly as billions of dollars flow into AI infrastructure. His distinction between AI model developers and data center operators challenges the prevailing narrative that all AI-adjacent investments are equally positioned for growth.
This matters because the data center boom has driven substantial market gains, with investors treating infrastructure providers as pure AI plays. If Chanos is correct that these companies function more like REITs than tech firms, their valuations may be fundamentally misaligned with their economic reality. The capital-intensive nature of data centers—requiring massive upfront investment in real estate, power infrastructure, and rapidly depreciating GPU hardware—creates a very different risk-return profile than software-based AI model companies.
The broader implication concerns AI capex sustainability. As companies like Microsoft, Google, and Meta pour tens of billions into AI infrastructure, questions about ROI become increasingly critical. If the anticipated AI revenue doesn’t materialize quickly enough to justify these investments, it could trigger a reassessment across the entire AI infrastructure sector, potentially impacting everything from chip manufacturers to power utilities serving data centers.