The AI building boom has a money problem, at least on paper. Gartner says more than $3.5 trillion of AI infrastructure spending could be exposed to impairment by the end of 2030 if AI companies keep spending at today's pace.

That is a projection, not a prediction that the money is already lost. But the numbers behind it are worth a closer look, because they show how thin the margin for error has become.

In plain words, an impairment is when a company admits that something it owns is worth less than the price on its books. The company writes down the value and takes a hit to its reported profit. No cash leaves the building that day, but investors learn the asset won't earn what was hoped.

According to Gartner, AI providers are expected to spend nearly $1 trillion on AI infrastructure in 2026 and another $6 trillion between 2027 and 2030. To cover that spending, the financing costs and a 12% return, data centre owners would need about $8.5 trillion in revenue through 2030, Gartner says. It forecasts that enterprise and consumer AI revenue tied to those data centres will reach no more than $3.6 trillion.

Gartner Newsroom: Gartner Says Over $3.5 Trillion of AI Infrastructure Spending Could Be Exposed to Impairment by the End of 2030 #GartnerNewsroom https://t.co/dbGFxPkQey

— Gartner (@Gartner_inc) October 14, 2026

Put those two figures side by side and the gap is stark. By Gartner's own math, expected AI revenue covers only about 42% of what the build-out needs to pay for itself, leaving a shortfall of roughly $4.9 trillion. Gartner also says less than half of AI revenue will go to companies that actually own the hardware, because many AI firms rent their computing instead of buying it.

The core issue is speed. AI chips are improving so fast that a server bought today can look dated within a few years. Gartner says spending on AI-optimised cloud computing will grow from $66 billion in 2027 to $174 billion in 2030, but that this is too slow for most providers to earn back their investment before the current generation of chips is replaced.

Big tech companies are already disagreeing about how long this hardware lasts. Amazon shortened the assumed life of some of its servers and networking gear from six years to five in 2025, citing the fast pace of AI development. That change alone added $889 million in depreciation costs in the first nine months of 2025. Meta went the other way, extending most of its servers and network gear to 5.5 years, which lifted its net income by $1.96 billion over the same nine months, according to its filing.

Same kind of hardware, opposite bets. If Amazon's view turns out to be closer to reality, companies using longer lifespans may be flattering their profits today and storing up write-downs for later.

Not everyone is equally exposed. Gartner says hyperscalers such as Amazon, Microsoft and Google have huge, varied businesses that could plausibly carry AI spending for years. Neoclouds, the newer companies that mainly rent out GPU capacity, have no such cushion. Gartner predicts that over 80% of them will stop operating independently by 2030, which points to mergers, takeovers or shutdowns.

Lenders and investors who funded those companies, often using the GPUs themselves as security, would feel that too. A chip that loses value quickly is weak collateral.

For ordinary users, a shake-out may not be all bad. Overbuilt capacity usually means cheaper computing, which can lower prices for AI tools and services. The risk sits more with jobs at smaller cloud firms and with retirement funds and shareholders heavily tied to AI infrastructure stocks.

Gartner's figures depend on its assumptions, including that spending continues on its current path and that 12% is the right return target. Revenue could grow faster than expected, or companies could simply slow their spending. Still, the report adds a well-known research firm to the list of voices asking whether AI's revenue can catch up with its bills.