An AI Crash Would Cost Stocks $20T, TS Lombard Estimates

In a crash scenario for the AI boom, equity holders would lose about $20 trillion and credit losses on AI debt would total $125 billion to $190 billion, according to TS Lombard’s Macro Picture “Wargaming the AI Crash” by Dario Perkins, dated 22 October 2026. We read four pages of the report that the account FinGenAi shared on X on 24 October 2026.

The report’s summary says: “An AI-related crash would have serious repercussions for financial markets and the global economy.” It adds: “But our analysis points to only a mild US recession, as lower rates quickly revive the forgotten bottom of America’s “K”.”

This Week In The AI Supercycle · VIDEO

Our weekly explainer of the AI supercycle, 18-24 October 2026. It covers the week’s debt and revenue stories and does not discuss TS Lombard’s report.

What the Scenario Says About Debt

TS Lombard writes that it is “pencilling in” early 2028 for the end of the boom, and that by then the hyperscalers “will have issued around $780 billion in debt securities.” It does not see them as the main credit risk: “We don’t think these companies are a serious credit risk.”

The strain, the report says, is “going to appear upstream,” among neoclouds such as CoreWeave, single-asset project vehicles and developers, Oracle, the standalone model developers (OpenAI, Anthropic) and AI app startups.

By the end of 2027, it estimates the non-hyperscaler part of the AI ecosystem will hold around $545 billion in bonds, $920 billion in bank loans and around $650 billion in private credit. It puts the total debt stock at around $2.9 trillion.

Source: TS Lombard, “Wargaming the AI Crash” (Dario Perkins, 22 October 2026), page 9, as shared on X on 24 October 2026. The lines carry different dates, as labelled.

What Was Posted on X

The @ackmeni post that shared the report summarises TS Lombard’s Wargaming the AI Crash as a $3 trillion revenue test: AI companies would need $2.5 trillion to $3 trillion in annual revenue by 2030, around 14 times today’s numbers. Those revenue figures are the post’s.

How It Gets to $125 Billion to $190 Billion

The report says the estimates were run with Claude. It assumes “almost 25%” of the debt stock is vulnerable, including the neoclouds, developers without firm hyperscaler leases, GPU-backed loans and lab-linked contracts.

It then applies “a cumulative default rate of 35% over the period 2028–30 (and a 25–50% recovery range),” which “gives overall credit losses of $125–190 billion for the industry overall.” It calls that “Material, for sure, but not exactly systemic for the US financial system as a whole.”

By our arithmetic, almost 25% of $2.9 trillion is about $725 billion of vulnerable debt, and the $125 billion to $190 billion range is about 4.3% to 6.6% of the total stock.

The report says: “Equity holders take the biggest hit (to the tune of $20 trillion) but there is no 2008-style cascade of credit losses and economy-wide bankruptcies.” By our arithmetic, $20 trillion is about 105 to 160 times the credit-loss range.

Jobs and Growth in the Scenario

On the page we read, the report puts the jobs created at around 500,000, a figure that could rise to 800,000 by the end of 2027; the page’s charts track data-centre construction jobs. If those jobs disappeared, it writes, “we would be looking at perhaps a 50bps rise in the US unemployment rate.”

It sums up the direct effects as “a 2%pts hit to GDP with 800k job losses,” and argues the stock-market wealth effect on spending is weaker than investors fear: “The problem with wealth effects is that there isn’t much evidence for them in the data.”

Why TS Lombard Doubts the Boom

The report’s case rests on where revenue comes from. TS Lombard writes that its sustainability metric, “the ratio of “internal” to “external” ecosystem revenues” is “pretty much the same” four months on, and that “this is still a boom about capex recycling and circular financing.”

On what investors should do, it says: “This bubble is not like Dotcom because it is much more concentrated on the US and the AI supply chain.”

The FinGenAi post that shared the pages also says AI companies would need $2.5 trillion to $3 trillion in annual revenue by 2030, around 14 times today’s level. That figure is the post’s; it does not appear on the four pages we read.

Connecting the Coverage

The question TS Lombard asks is the one in our Ramp piece of 23 October: how much revenue the AI ecosystem earns from outside itself. That piece reported Ramp’s finding that “80% of spend on OpenAI and Anthropic is driven by 1% of firms” on its platform.

On debt, Bloomberg reported on 23 October that chip-backed debt is on track to become a trillion-dollar market, as our chip-debt piece covered. Our commitments tracker counted about $3.56 trillion of off-balance-sheet commitments at eight large companies, by our arithmetic from their filings.

Separately, as our selloff piece reported, the 22 October selloff in AI stocks had reached bonds, credit spreads and gold by the next morning. CoreWeave, which TS Lombard names among the more exposed borrowers, still needs new debt, with Goldman Sachs and Bernstein far apart on how much, as our CoreWeave piece covered.

What is new here: a research house putting numbers on a crash scenario, with the larger losses on the equity side, as the report puts it.

The Structural Read

As we read it, the report locates the risk outside the hyperscalers: it says they are not a serious credit risk and that the strain would appear upstream, among neoclouds, project vehicles, Oracle, the model developers and AI app startups.

As we read it, the scenario rests on a revenue test: TS Lombard says its ratio of internal to external ecosystem revenues is pretty much the same four months on, which is why it calls this a boom about capex recycling and circular financing.

As we read it, the macro call is milder than the market call: the report sums up the direct effects as a 2%pts hit to GDP with 800k job losses and calls the outcome only a mild US recession.

Three Implications

FOR LENDERS The report treats about 45% of non-hyperscaler private credit and 25% of bank loans as vulnerable in its scenario.

FOR EQUITY HOLDERS The report puts their hit at about $20 trillion, the largest loss in the scenario.

FOR THE TIMELINE TS Lombard says it is pencilling in early 2028 for the end of the boom, and writes that it does not know when the boom will end.

The Business Engineer Lens

This story maps onto the Business Engineer framework The AI Supercycle.

The framework puts it this way: “Layer 5 — Compute Capacity. Hyperscaler GPU clusters, neoclouds, custom-silicon estates. Where the capex bubble lives.”

As we read it, part of TS Lombard’s scenario sits on that layer: the neoclouds, such as CoreWeave, and the single-asset project vehicles that the report names among the borrowers where the more serious strains would appear. The framework does not discuss this report.

What Is Not Established

We read four pages of the report (pages 1, 3, 9 and 11), as images shared on X, not the full report or TS Lombard’s own release. This is a scenario, not a forecast that a crash will happen; TS Lombard writes that it does not know when the boom will end. The $2.5 trillion to $3 trillion revenue figure comes from the post that shared the pages, not from the pages we read. We did not contact TS Lombard.

The Bottom Line

TS Lombard’s crash scenario puts AI-related debt near $2.9 trillion and the credit losses at $125 billion to $190 billion, against a $20 trillion hit to equity holders and what it calls only a mild US recession.