Gartner: more than 3,500 billion dollars’ worth of AI infrastructure may need to be written down by the end of 2030

According to the consultancy, the expected revenue from AI will not cover investment in data centres; more than 80 per cent of ‘neoclouds’ are expected to lose their independence by 2030.

In an analysis published on October 14, 2026, Gartner quantifies the gap between the wave of investments in AI data centers and the revenues they are expected to generate. To cover their capital expenditures and financing with a return on invested capital of 12%, these infrastructures would need $8,500 billion in cumulative revenues by 2030; the firm forecasts only $3,600 billion, less than half of which would go to infrastructure providers. The result: more than $3,500 billion in assets exposed to depreciation by the end of 2030, and more than $4,000 billion by 2032.

Neoclouds on the front line

The shock would be unevenly distributed. Hyperscalers can absorb longer payback times thanks to their other activities, while specialized GPU rental companies, known as "neoclouds", are much more fragile: Gartner estimates that more than 80% of them will have stopped operating independently by 2030. The firm estimates that AI infrastructure will spend around $1,000 billion in 2026 and $6,000 billion in 2027 to 2030.

It recommends that cloud buyers require proof of funded capacity, investors test their models with different lease prices and hardware lifetimes, and operators tie their commitments to verified demand rather than industry forecasts.