Gartner warns $3.5tn of AI infrastructure spending is at write-down risk

“All of them want to be the full-stack. They can’t all be,” Gartner analyst Will Sommer told Bloomberg .

More than $3.5tn of AI infrastructure spending faces possible impairment by the end of 2030, Gartner said on 14 October. That would happen if AI providers keep spending at their current pace. Companies must write down impaired assets, because those assets can no longer earn back their cost.

The research firm also predicts that more than 80% of neoclouds will stop operating independently by 2030. Neoclouds are companies that rent out computing power built for AI. Gartner blames cost pressure and the lack of a business model that sets them apart.

Gartner expects AI providers to spend nearly $1tn on AI infrastructure in 2026, Sommer, a senior director analyst, said in its statement. It forecasts another $6tn between 2027 and 2030.

Through 2030, AI data centre owners would need $8.5tn in revenue to cover that spending and its financing costs. That figure also includes a 12% return on invested capital, which Gartner assumes to be the hurdle rate for investment. Gartner forecasts no more than $3.6tn of business and consumer AI revenue tied to AI data centres in that period.

“Our research indicates that enterprise and consumer AI revenue will be insufficient to support profitable returns on all planned AI capital expenditures during this period,” Sommer said.

Most AI-native providers run asset-light businesses. So Gartner expects less than half of AI revenue to go to providers that own significant AI infrastructure.

If spending stays on its current path, more than $4tn in asset value could face write-down risk through 2032. In that case, Gartner does not expect the 12% return to arrive until the end of 2032 at the earliest. Sommer, Mark McDonald and John Lovelock wrote the research note behind the figures.

Hyperscalers have large and varied revenue that could support AI spending for years, Gartner said. Neoclouds must soon earn returns on AI-optimised cloud infrastructure to keep running.

Gartner expects spending on that infrastructure to grow from $66bn in 2027 to $174bn in 2030. That growth is too low, Sommer said. Most AI compute providers will not earn back their investment before newer AI chips replace the current generation.

Bloomberg named CoreWeave and Nebius as specialist providers more directly exposed than Microsoft, Alphabet and Amazon. CoreWeave’s shares and bonds fell steeply this week over its AI debt.

Many neoclouds want to sell the higher-margin software and services of cloud providers, Sommer told Bloomberg. Few may succeed.

“For every dollar that’s spent on compute, only a fraction of that is going to the owners of compute,” Sommer said.

“Not everyone is in trouble. Some will outperform, some will underperform, and some will have to exit the market,” Sommer told The Register.

Gartner’s numbers follow a Bain & Company estimate that the AI industry needs $6tn in annual revenue by 2031. This week, the IMF also warned about circular AI deals .