Steve Eisman isn't calling the AI boom fake. He's saying investors are paying up for the companies writing the checks, when the cleaner trade may be the companies cashing them.
If you're buying the AI boom by simply buying the biggest cloud names, Eisman's warning is hard to ignore. The Neuberger Berman portfolio manager, best known for betting against subprime mortgages before the 2008 crash, told Fortune that investors are crowding into the wrong side of the trade: the hyperscalers spending hundreds of billions of dollars to stay in the race.
His case is not that AI fails. That would be too easy, and frankly too lazy. His point is sharper. Microsoft, Alphabet, Amazon, Meta Platforms, and Oracle are pouring money into data centers, chips, leases, and power before the returns are fully visible. Fortune reported that Eisman sees aggregate AI capital expenditure across the major cloud providers moving from roughly $400 billion last year toward close to $1 trillion in 2026. You don't need to believe every dollar of that forecast to see the pressure building.
The numbers are already uncomfortable. Investors Business Daily, citing Yardeni Research, recently put Alphabet's 2026 capital spending forecast at $180 billion to $190 billion, more than double its $91 billion in 2025. Business Insider reported earlier this year that Amazon laid out a roughly $200 billion 2026 spending plan. Axios reported this month that Alphabet, Amazon, Meta, Microsoft, and Oracle had already raised $255.34 billion through stock and debt in 2026, while Barron's put their planned AI data center spending by year-end near three-quarters of a trillion dollars.
That money has to come from somewhere. Sometimes it comes from operating cash flow. Sometimes it comes from debt. Sometimes it comes from issuing new shares. None of those routes is free for you as a shareholder.
Eisman's analogy is blunt enough to be useful. He has compared the cloud giants to airlines: businesses that can be essential, huge, and still brutal for investors because they require constant capital and face hard competition. If you want to own that, fine. But don't pretend you're buying the toll booth when you're really buying the company paving the road.
The suppliers get paid first
Eisman's preferred exposure sits one layer below the hyperscalers. He has pointed to semiconductors, networking equipment, and alternative energy, the companies selling what the cloud giants need to build. Nvidia is the obvious chip name. Arista Networks and Cisco sit in the networking layer. Power providers and grid equipment companies enter the discussion because AI data centers don't run on investor excitement. They run on electricity, lots of it.
This is the old picks-and-shovels trade, but the timing matters. The market has spent much of the AI boom treating Amazon Web Services, Google Cloud, and Microsoft Azure as the safest way to own the theme. Eisman is saying safety and leverage are different things. A supplier can benefit from the spending race without having to prove that every application built on top of that infrastructure earns back the money spent underneath it.
Look at the recent market action. The Guardian reported that the Nasdaq fell 2.2% on June 23 as a tech and AI selloff shook markets from Wall Street to Asia. Axios reported the next day that the Nasdaq 100 had slid 3.3% and the S&P 500 had fallen 1.4% after chip-heavy selling in South Korea rattled investors. Earlier in June, the Times of India reported that chip stocks including Nvidia, Micron, and Broadcom lost about $1.3 trillion in market value in less than 24 hours. These aren't tiny moves. They are the market testing how much confidence is already baked into the trade.
Goldman Sachs has been circling the same issue from another angle. MarketWatch reported this month that Goldman strategists warned hyperscaler capital expenditure estimates keep rising while AI-related valuations remain stretched. Another Goldman strategist, Rich Privorotsky, described a widening split between the hardware companies that benefit from spending and the hyperscalers whose stocks have lagged as their commitments have increased.
Here's the thing: Eisman doesn't need AI to disappoint for his argument to work. He only needs the cloud companies to spend too much, too quickly, in a market where model performance gets copied, compute prices fall, and customers switch tools more easily than investors want to admit. If that happens, the suppliers can still have a good business while the spenders fight over returns.
That is the part investors should sit with. Nvidia doesn't need Microsoft to win every enterprise AI account. Arista doesn't need Amazon to prove every new data center earns an attractive return. Cisco doesn't need one model to become the permanent standard. The suppliers need the buildout to continue, and for now it is continuing at a scale that would have sounded absurd before ChatGPT turned cloud infrastructure into the most expensive arms race in technology.
Eisman's track record doesn't make him automatically right. It does make the warning harder to dismiss. He isn't calling for an AI collapse. He's calling out a mispricing inside a real boom, and those are the calls that hurt most when investors ignore them.
Also read: Satya Nadella says companies that rent their AI brains are making a strategic mistake they will regret | NVIDIA has taken the top spot in datacenter Ethernet switching and it now controls the full AI stack | Oracle cut 21,000 jobs in a year and put the blame directly on AI
