Chris Wood has spent years writing an investment letter called GREED & fear, and the title has seldom fit a quarter as neatly as it fits this one. In his Asia Maxima note published this week and titled “Reverberating tremors,” the Jefferies global head of equity strategy compared the artificial-intelligence boom to the subprime mortgage machine that came apart in 2008, and argued that much of the money now pouring into American data centers will end up destroyed.
The comparison is not about the technology. It is about who is paying for it. For the first three years of the AI build-out, the largest cloud companies funded their data centers out of the cash their existing businesses produced. That is changing. Wood reckons hyperscaler AI capital spending will approach $1 trillion next year, up from roughly $700 billion this year, and a growing share of it is being borrowed rather than earned.
“Media reports and sell-side research persist in claiming no bubble exists,” Wood wrote, according to accounts of the note. “Negative news flow will grow as the year progresses.”
His sharpest point concerns the small group of customers standing behind the spending. Nearly half of a $2.34 trillion backlog of contracted orders owed to the big cloud providers traces to OpenAI and Anthropic, two labs that do not yet make a profit. If those companies stumble, analysts said, the obligations do not disappear; they land on the balance sheets of the companies that signed them.
Then there is the debt that does not show up in the headline numbers. Data-center leases and other long-term commitments are increasingly kept off balance sheet, which means the true obligations of the largest technology companies run well above what their reported accounts suggest. Wood has pointed to estimates of hidden liabilities running into the hundreds of billions of dollars, a figure large enough to shift how bond investors price the sector.
The quality of earnings has drawn similar scrutiny. A meaningful slice of the profit growth at the companies riding the AI trade comes from paper gains on equity stakes in AI startups rather than from cash-generating operations. Those marks rise when valuations rise, and they can fall just as quickly. Meanwhile, operating cash flow at several of the biggest spenders is shrinking even as they plan to accelerate capital expenditure.
Wood’s base case is blunt. Asked in September what he expected from the spending wave, he said the hyperscalers would probably fail to earn a return that justifies the investment. “My base case remains that they will not generate returns sufficient to justify those investments,” he said. “Ultimately there will be significant capital destruction.”
He also expects the spoils to migrate. Cheaper open-weight models from Chinese labs, he argued, will take share from American providers that charge more for similar capability, pressing down the prices those companies can charge and, by extension, the returns they can book.
The strategist is not the first to sound this alarm, but he is among the most closely followed. His reputation rests on calling the 2008 crisis and, more recently, on steering clients into Asian equities while Wall Street chased American growth. Analysts said the timing of the note matters: it arrives as investment-grade bond issuance from the cloud companies has scaled up sharply, turning them into competitors for the same pool of credit that funds the US Treasury.
That is where the cycle turns, in his telling. “The moment credit markets effectively withdraw funding, the entire AI trade could unravel,” he said. Semiconductor shares have carried the market higher for two years, and Wood suggests the easy money in those stocks has already been made. He now prefers the companies that supply the industry with chips over the companies borrowing to build with them.
He reserves a measure of suspicion for another group: the executives who have publicly called for a pause in frontier development. Such calls, he suggested, may serve less as caution than as an effort to raise barriers that entrench incumbents. A slowdown that freezes the leaders in place is a slowdown that protects them.
None of this amounts to a prediction of imminent collapse. Wood’s warning is about the shape of the ending, not its date. The AI trade, he said, will hold together as long as credit keeps flowing and buyers keep paying the premium. What would break it is not a failed model or a disappointing product cycle but a change in the willingness of lenders to fund the next round of concrete and copper.
For investors, the practical question is not whether the technology matters. It plainly does. The question is who ends up owning the bill when the borrowing outruns the returns. Wood’s answer is that the capital already committed will not all come back, and that the reckoning will arrive through the credit markets rather than the earnings report. Until then, the spending continues, and so does the argument about what it is really worth.
