What Happens If the AI Trade Cools? A Scenario Map
Nobody can time the moment capital gets more expensive. It is still possible to describe what happens first, what happens second, and who feels each stage.
Every technology cycle that attracts this much capital eventually faces a period in which the capital becomes more expensive. Nobody can time that transition, and the people who claim to are usually selling something. But the sequence of effects is describable, and describing it is more useful than dating it.
What breaks in which quarter
- Late-stage private valuations mark down first, because they are repriced by negotiation rather than by trading.
- Talent-only acquisitions accelerate as companies lose the ability to raise independently.
- Capital expenditure plans are stretched rather than cancelled, because the facilities are already committed.
- Model releases slow, and the surviving labs prioritise efficiency over capability.
- Open-weight models gain share, because economic pressure makes the free option attractive.
Labs with no distribution, developers with no tenants
The most exposed businesses are those whose value depends entirely on a growth narrative: early-stage labs without distribution, data providers serving only frontier research, and infrastructure developers funded on the assumption that future tenants will arrive and sign.
The least exposed are the ones with existing revenue and a cheap supply of capital. They would see weaker demand but retain the ability to invest while competitors cannot, which is historically how consolidations produce winners — the buyer sets the price when nobody else can bid.
The demand that does not go away
The case against a correction is that the underlying demand is real. Organisations are paying for these systems because they do something useful, and the cost of useful capability keeps falling. A business with real demand and falling unit costs can survive a great deal of enthusiasm evaporating around it.
Both things can be true at once: the technology can be transformative and the current set of valuations can be wrong. That was the situation in every previous cycle, and there is no reason to expect this one to be tidier.
The systems keep running either way
Even in a severe correction, the deployed systems would keep running. Organisations that have integrated a model into a workflow do not un-integrate it because the vendor’s valuation fell, and support contracts rarely have a clause for that. The installed base is real, and it generates revenue regardless of what happens to growth expectations.
Research would also continue at the largest labs, since their budgets are funded from existing cash flows rather than from the capital markets. What would change is the number of new entrants and the appetite for pre-revenue companies.
The dot-com template, with the fibre still in the ground
The dot-com correction destroyed a large share of the companies funded during the boom and left the infrastructure and the surviving businesses in place. Capacity that had been overbuilt was absorbed over time, often at pennies on the dollar, by the companies that survived.
That is the most likely shape here as well: substantial losses for late-stage investors, a wave of consolidation, and fewer participants running more durable economics on equipment somebody else paid for.
Image credit and licence details for every photograph on this site are listed on the credits page. This article is editorial content; it carries no sponsored material.