A Wall Street Firm Modeled Out an AI Crash: S&P 500 Down 30%, $20 Trillion Gone. But Here's What Happens Next
Tech stocks
closed out October on Friday with the Nasdaq Composite's worst month since March 2025, and the strain has reached the bond market too. Intel(NASDAQ:INTC) stock plunged 29% in one day, and CoreWeave(NASDAQ:CRWV), whose bonds came under heavy selling over worries about an AI slowdown and its weak balance sheet, has seen its stock drop 35% over the month.
Those jitters are the backdrop for a recent report in which TS Lombard's Dario Perkins models what happens if the AI investment boom ends. In his scenario, U.S. stocks fall at least 30%, wiping out about $20 trillion in market value. However, the firm's advice is still calm: hold a normal amount of risk through diversified global stocks, don't bet against tech too early, and add government bonds only once strain shows. The firm's working assumption shows a reversal in early 2028, though it says that date could slip. Perkins writes that the recent market wobble may be the beginning of the end or may blow over, and the report does not claim to know which.
Image source: Getty Images.
Missed AI's "Act 1"? Act 2 Could Be 14x Bigger. Most investors think they missed the AI boat because they didn't buy Nvidia in 2005. But according to our analysts, we're only at the end of "Act 1"—the R&D phase. "Act 2" is the global rollout. Continue »
The one move: stay invested, stay global, wait for strain
Perkins, managing director for global macro at TS Lombard, says the firm does not know when the boom will end, so its answer is not to try to time it. In the report's words, "Investors can maintain a normal amount of risk via a diversified portfolio of global stocks." The firm argues this bubble differs from the dot-com era because it is "much more concentrated on the U.S. and the AI supply chain," so investors can hold exposure that sidesteps the most concentrated bets.
Bonds come second. TS Lombard says investors "can also add government bonds, but only once signs of fragility emerge." The reasoning comes down to timing. Shorting U.S. tech now, or adding long-dated bonds early, would risk underperforming if the investment boom runs for several more years, with stocks rising and yields drifting higher in the meantime.
The report also expects non-U.S. stocks to outperform over time and argues the dollar is ultimately headed lower.
What a crash could look like
In the scenario, the S&P 500 falls at least 30%, with tech down 40% and other sectors down 20%. That erases about $20 trillion of market value. The report's detailed estimate puts losses on AI-related debt, including private credit, at $125 billion to $190 billion. Perkins said in the report that "the AI crash scenario is now overdetermined. There is just a lot that could go wrong."
The weakest links are AI cloud providers such as CoreWeave, Oracle(NYSE:ORCL), stand-alone model developers like OpenAI and Anthropic, and construction and equipment suppliers. Hyperscalers, the giant spenders, would face revenue downgrades but are not seen as serious credit risks.
Even so, TS Lombard expects only a mild U.S. recession, comparable to 1990 or 2001. That's because the Fed has room to cut rates, private-sector balance sheets are healthy, and rate cuts would help the lower end of what the report calls America's K-shaped economy. It estimates the crash would cut about 2 percentage points from the level of U.S. GDP. It also estimates that the AI data-center construction boom has created around 500,000 jobs, which could reach 800,000 by the end of 2027.
The financing worry, in plain English
The report's central concern is how the buildout is paid for. AI labs sign contracts promising to pay for computing power they haven't yet used. The cloud provider or hyperscaler counts that promise as backlog and borrows against it to build the data center. TS Lombard likens the labs to the no-income borrowers of the subprime era: the data center is the collateral, and the structure holds only while the lab can keep raising money. The firm calls the subprime comparison a little stretched and says the crash would be nothing like 2008.
The numbers behind the worry are large. U.S. tech companies are investing around $1 trillion a year, and AI companies are expected to issue around $700 billion in new bonds in 2027, $400 billion of it from hyperscalers. Hidden liabilities from off-balance-sheet structures are estimated at $1.5 trillion to $2 trillion, and CoreWeave is already paying 15% interest. Revenue is still largely recycled spending, and 80% of OpenAI's and Anthropic's combined earnings come from just 1% of their customers, mostly unprofitable AI start-ups.
The long view
For investors thinking in years rather than months, the report's conclusion is more reassuring than its headline numbers. In TS Lombard's model, most of the damage lands on stock prices, not on the financial system: credit losses are real but not systemic, and the firm sees no 2008-style wave of defaults spreading across the economy. "We're not expecting major job losses and the Fed has plenty of room to cut interest rates," the report says.
Cheaper borrowing could also unlock demand that has built up in corners of the economy squeezed by tight monetary policy, housing among them. "If lower rates can unlock that potential, it should be possible to deliver a swift economic recovery," the report says. The firm expects stock markets outside the U.S. that largely missed the AI boom to outperform, perhaps significantly, and expects long-term interest rates to fall. That falling-rate outlook is the reason it wants government bonds in the mix once strain shows.
That logic sits behind the playbook. A diversified spread of global stocks should do well if the boom keeps running and should beat a concentrated bet if it breaks. And as the TS Lombard report puts it, "If something can't go on forever, it will stop" -- but even in the model, the end of the boom is a recession to sit through, with a recovery on the other side.
Missed AI's "Act 1"? Act 2 Could Be 14x Bigger
Most investors think they missed the AI boat because they didn't buy Nvidia in 2005.
But according to our analysts, we're only at the end of "Act 1"—the R&D phase. "Act 2" is the global rollout, a massive technology build-out that Google's CEO says could be more profound than fire or electricity.
While everyone watches the chip designers, one "backbone" company provides the critical technology that Nvidia and AMD literally cannot function without. We've detailed this under-the-radar player in our new AI Playbook.
Jonathan Ponciano has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Intel and Oracle. The Motley Fool has a disclosure policy .
A Wall Street Firm Modeled Out an AI Crash: S&P 500 Down 30%, $20 Trillion Gone. But Here's What Happens Next was originally published by The Motley Fool
