Bill Eigen isn't calling the AI trade a bubble. He's calling it something more specific: a real estate cycle wearing a tech costume.
The J.P. Morgan Asset Management manager told CNBC he's less worried about how fast AI revenue is growing and more worried about the rate at which that growth is accelerating, or decelerating. He calls it the second derivative, and he said both AI capital spending and private market valuations of AI labs already appear to be slowing down.
The Number That Actually Scares Him
"What I'm terrified of is when that starts to slow," Eigen said.
That fear has real scale behind it. Hyperscaler capital spending is projected to reach roughly 3.1% of U.S. GDP by 2027, according to Apollo Global Management chief economist Torsten Slok, more than double the 1.2% peak the telecom sector hit during its own buildout boom two decades ago.
Why He Called It a Real Estate Cycle
The comparison comes from how the debt is structured, not from the technology itself.
Eigen said he's wary of long-dated debt tied to data centers because of a mismatch: bonds financing these projects often run 30 years, while the chips inside the buildings depreciate over just three to six years. That's a classic real estate financing problem, borrowing long against an asset that ages fast.
Much of that obligation doesn't show up on balance sheets at all. Goldman Sachs estimates hyperscalers carry roughly $1.5 trillion in aggregate lease commitments structured to stay off their books, including Meta's $27 billion Hyperion joint venture with Blue Owl. A separate Wall Street Journal analysis found Alphabet, Amazon, Meta and Microsoft have collectively racked up $3 trillion in commitments outside their reported balance sheets.
Credit spreads across public markets remain near historic tights. But Eigen pointed to credit default swaps on AI-related companies, including Nvidia, which have started widening even as headline valuations haven't moved to reflect it.
That divergence is the part Eigen finds most telling. Public equity prices are a lagging signal in a cycle like this, he suggested, while the credit market, where investors are pricing actual default risk rather than growth optimism, tends to move first. Widening spreads without a corresponding stock price move is exactly the kind of early warning a real estate investor would recognize from prior cycles, long before headline valuations catch up.
Not Buying, Not Selling
Eigen said he isn't shorting AI assets, since timing the end of a cycle like this is close to impossible. He's also not buying at current prices.
Part of his hesitation comes down to roughly $2 trillion in remaining performance obligations that major hyperscalers have reported, representing revenue they've contracted for but haven't yet collected.
"How's that going to get paid?" Eigen said.
That question sits underneath most of the AI infrastructure buildout right now, including deals like Nvidia's recent $105 billion financing commitment for OpenAI's Ohio data center and the wave of neocloud financing rounds following it.
What This Means for Miami
Eigen's real estate framing should land differently in Miami than almost anywhere else. This is a market where family offices, private wealth managers and real estate investors already think fluently in terms of duration mismatch, leverage structure and off-balance-sheet obligations, the exact vocabulary Eigen is now applying to AI.
For Miami's investment community evaluating exposure to AI infrastructure, whether through public equities, private credit, or direct stakes in data center projects, Eigen's framing offers a more familiar due diligence checklist than most AI coverage provides. Ask about lease terms and depreciation schedules the way you'd underwrite a commercial property, not just about revenue growth, and the risk picture looks considerably different than the headline numbers suggest.
That's a particularly relevant lens given how much AI infrastructure activity is now touching South Florida directly, from data center development in Miami-Dade to the region's growing role as a hub for the private capital funding these projects nationally. Local investors don't need to guess whether this cycle resembles something they've seen before. For many of them, it's the same real estate math they already know, just with GPUs instead of buildings as the depreciating asset.