$1 Trillion in AI Commitments Rests on $200 Billion in Revenue

TS Lombard says the AI crash is coming — and the firm isn't interested in guessing the date. In a recent client report, global market strategist Dario Perkins laid out a fictional, multiyear timeline of the bubble bursting and warned that positioning matters far more than timing.

"The only question is what to do about it," Perkins wrote.

The report nods to Citrini Research's viral AI doomsday piece from the start of the year. Perkins picks up the format and writes from the vantage point of an investor in 2030, looking back on 2026 as the year things started to look rotten below the surface.

What does Perkins' fictional timeline show?

In his 2026 scenario, hyperscalers were flooding the AI ecosystem with cash while Nvidia led a wave of circular investing — putting money into its own customers to drive chip purchases. Hyperscalers started funding build-outs through special-purpose vehicles kept off balance sheets, and leverage began to explode. Model developers Anthropic and OpenAI stayed secretive about their finances, and concentration risk kept building.

"The entire trade was an act of faith, with just $200 billion in current revenues supporting more than $1 trillion in future commitments from just two companies," Perkins wrote. That faith persisted even as the public showed a lack of interest in AI.

The vibe shifted in late 2027, in his telling. OpenAI's much-delayed IPO didn't get the benefit of the doubt that Anthropic's offering got a year earlier, because investors had started reading the numbers closely.

"Meanwhile, people were starting to think about the 2028 U.S. elections and the one issue that did unite Americans — anti-AI sentiment," according to Perkins. "It was clear a political/regulatory backlash was coming."

How bad is the imagined crash?

In the scenario, the bust starts in early January 2028. Hyperscalers stop investing and slash capital-expenditure guidance by 45%, to around $600 billion. Stocks fall 10% in a week.

"For everyone downstream, lower hyperscaler capex became a revenue cliff. Orders at chip, server, construction and power-equipment suppliers suddenly halved, and the export-led economies that had supplied them — Korea and Taiwan — were hit hard," he wrote.

The numbers accelerate from there. By April 2028, the S&P 500 sits 30% below its peak, AI-linked stocks are down 40%, the rest of the market is down 20%, and $20 trillion of U.S. financial wealth has evaporated. AI technology itself is freely available by then — and, in Perkins' words, "most regular people weren't that fussed."

The bear market brings the cleanup: investors grill AI balance sheets, neoclouds file for bankruptcy, data-center loans take write-downs, a cloud provider gets cut to junk and banks are left holding debt. Hyperscalers become slower-growth assets and some lenders face billions in losses.

What happens to the economy after the pop?

The fallout, in Perkins' scenario, is real but contained. The AI bust wipes out 500,000 jobs in nonresidential construction, and Asian supply chains take heavy damage. A recession follows — a mild one by historical standards, because AI debt isn't systemic and defaults stay mostly tech-related. Central banks start cutting rates aggressively by mid-2028.

Those cuts turn housing, residential construction and consumer durables around within a few quarters. The stock mania gets replaced by a housing boom. Businesses eventually figure out how to use AI productively, but it doesn't replace labor. The technology becomes "highly commoditized, thanks to open-weight models that made intelligence cheap, squeezed margins at the model layer and shifted the value to infrastructure, applications and users."

What should investors actually do?

Don't try to time it, Perkins says. Exiting AI investments too early could mean losing money for years. "The canary in the coal mine is not the problem, it's the mine itself that is," he said.

Hedge fund manager Ken Griffin, who owns market maker Citadel Securities, recently echoed the stance — saying AI stocks are probably in a bubble but still worth holding.

Perkins' advice is deliberately modest: hold a normal amount of risk, protect the downside, and run a diversified portfolio of global stocks with less of the AI complex than the index gives you.

"This is not like dot-com, which spread across the developed world, Europe included," he added. "This time, the bet is concentrated in U.S. tech and an Asian hardware supply chain, so most of the rest of the world should fare better — eventually. And add duration as the vulnerabilities appear, because bonds are what rallied."

The bottom line: TS Lombard treats the bust as a base case, not a tail risk — and the trade now is portfolio construction, not a date on the calendar.