RiskReversal Pod Sep 25, 2026 1h 36m · 80m saved
With Jim Chanos, Founder of Chanos & Company · Gary Marcus, Professor Emeritus at NYU and author of the Marcus on AI Substack
Jim Chanos said returns on the new money the big cloud companies are putting into AI peaked in 2024, and that on the current rate of decline they would fall below those companies' cost of capital sometime in mid-2027.
Most investors treat the AI build-out as a technology story. Chanos approaches it as a financing story, and he said the thing that ends it will be the capital markets, not the technology.
"There is very big risk of financing in this market that will make all of this reflexive and will look like a crash where long-term secular demand could still be growing."
Jim Chanos, founder of Chanos & Company, on the RiskReversal Pod, speaks as the short seller the host credited with first identifying the fraud at Enron. He now teaches a course on the history of financial fraud at Yale and the University of Wisconsin, and he appeared alongside AI researcher Gary Marcus.
The full interview is covered here so you can skip it. 96 minutes of audio, 16 minutes of reading.
Here are the 14 arguments that matter.
Key Takeaways
AI data centers are an equipment leasing business, in Chanos's reading, not a technology business
He said investors are wrongly treating today's spot prices as long-term prices
Hyperscalers' returns on incremental capital peaked in 2024 and could drop below their cost of capital by mid-2027
Convertible bonds make the neoclouds' borrowing look cheaper than it is
Credit markets have tightened for three to five months while stock markets have not
Nvidia is cheap next to the companies that depend on it, so he would be short its customers and long Nvidia
The internet did not raise US GDP growth in the decade after Netscape, and S&P 500 earnings grew more slowly
Microsoft and Meta are the healthiest hyperscalers; Amazon and Oracle trail
In 2000 and 2001, S&P 500 earnings fell 40% in about four quarters, and he said this boom is bigger
All four classic boom drivers are active at once: credit, government, technology and real estate
The Marcus proposal to shut down OpenAI fits his definition of fraud, Chanos said
1. Data Centers Are Leases
Chanos's firm first looked at the older, pre-AI data center companies in 2022, when they were structured as real estate investment trusts. He said those businesses earned low- to mid-single-digit pre-tax returns on capital even while cloud computing grew fast, because equipment broke, needed replacing and required backup. Marketing them as REITs let investors overlook the depreciation.
He can't judge the AI models, but he can judge a property deal
"But I can analyze a real estate deal. And that's where I think people are really jumping off the cliffs right now because they're assuming spot pricing is long-term pricing." — Jim Chanos
The business is buying chips and renting them out
"And what they don't realize is that the AI data center business is primarily an equipment leasing business. I'm buying chips from Nvidia and I'm renting them to you. And that's a financial construct." — Jim Chanos
He said he knew early on that hosting AI would become a commodity business, and said he would defer to Marcus on the models themselves.
