The investor who called the 2008 housing collapse now argues the most crowded AI positions are also the most dangerous, and the real money is in the suppliers, not the spenders.
Steve Eisman has a habit of being right when everyone else is comfortable being wrong. In 2008 it was mortgage-backed securities. Now, in a Fortune interview published today, the Neuberger Berman senior portfolio manager is making another contrarian call: most investors buying into the AI boom are buying the wrong companies, and they don't fully understand what they own.
His argument is straightforward. Hyperscalers like Microsoft, Alphabet, and Amazon are locked in a spending race that is outpacing their ability to generate free cash flow. As Fortune reported, Eisman estimates that aggregate AI capital expenditure across the major cloud providers is climbing from roughly $400 billion last year toward close to $1 trillion in 2026. The numbers are already visible in company guidance: Alphabet alone has flagged capital expenditures between $175 billion and $185 billion for the year, while Amazon has penciled in approximately $200 billion. When you stack Microsoft, Meta Platforms, and Oracle on top, aggregate hyperscaler capex for 2026 exceeds $452 billion. That money has to come from somewhere, and increasingly, Eisman argues, it's coming from public shareholders through dilution.
His analogy is blunt. He compares cloud giants to airlines: capital-hungry businesses with little pricing power and brutal competitive dynamics. If you want to own airlines, fine. But don't confuse that with owning the companies that sell Boeing its engines.
Eisman's preferred exposure sits one layer down from the hyperscalers, in the companies that sell them what they need to build. He flagged three sectors specifically: semiconductors, networking equipment, and alternative energy. These businesses capture hyperscaler capex without carrying it. Nvidia is the obvious name in chips, and Eisman has remained bullish on it even as he turns skeptical on its biggest customers. Arista Networks and Cisco sit in the networking tier he favors. On the energy side, the logic is simple: AI data centers are voracious power consumers, and whoever builds and supplies that power infrastructure benefits from every dollar of hyperscaler spending, regardless of which model wins the market.
The picks-and-shovels framing isn't new, but Eisman's application of it cuts against the consensus in a specific way. Most retail and institutional money flowing into the AI trade has landed on the hyperscalers themselves, treating Amazon Web Services or Google Cloud as the safest, most legible AI bet. Eisman's point is that legibility is not the same as leverage. AWS grew revenue 28% year over year in Q1 2026 to $37.59 billion, the fastest pace in 15 quarters, and the growth trajectory is real. But that growth is being funded at enormous cost, and whether AI monetization can justify $200 billion in annual capex remains genuinely open.
Benzinga noted that Eisman's skepticism extends particularly to Alphabet and Microsoft, two names that have functioned as de facto AI proxies for a broad swath of institutional portfolios. His concern isn't that the underlying technology fails. It's that these companies are competing in what amounts to a commodity arms race where switching costs are low and pricing power may never fully materialize.
The June market action lends the argument some urgency. A two-session AI selloff on June 24 and 25 wiped more than $1.3 trillion in semiconductor market value and sent the Nasdaq down 2.21% in a single day. Whether that was a correction or a signal is still being debated, but it arrived at a moment when Goldman Sachs equity research head James Covello had already warned publicly that AI monetization timelines were running behind spending timelines. His read: "At some point, you've got to make money."
Concentration risk hiding in plain sight
What Eisman is really flagging is a structural mismatch between where capital is concentrated and where the durable economic advantage actually lives. The hyperscalers are, in his framing, competitors in a market with no clear winner yet, all funding each other's buildout while suppliers collect. Nvidia doesn't care whether Microsoft or Amazon wins the enterprise AI contract. Arista doesn't care which model runs on the other end of the fiber. That indifference to the outcome of the competitive fight is exactly the kind of position Eisman prefers.
It rhymes with a rotation already underway among some institutional allocators, though not everyone is reading the setup the same way. David Tepper recently trimmed Nvidia by 13% and AMD by 32%, rotating toward hyperscalers directly, essentially the opposite of Eisman's move. The divergence matters: two credible investors, same data, opposite conclusions about where the leverage sits in the stack.
Eisman's track record earns his read serious weight. He isn't calling an AI bust. He's calling a mispricing within a real boom, which is historically the more precise and more actionable kind of contrarian call. The investors who made money in the California gold rush weren't the ones panning the rivers. They were selling the picks, the shovels, and the denim.
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
