Gartner's latest forecast arrived with the largest number in the AI infrastructure debate so far. More than $3.5 trillion of spending could be exposed to impairment by the end of 2030. The sentence that does the work sits right behind it: the exposure exists if providers keep spending at the current pace. This is a conditional scenario built from Gartner's own assumptions, not a loss anyone has booked.
The gap, stated plainly
Gartner expects AI providers to spend nearly $1 trillion on infrastructure in 2026 and another $6 trillion during 2027–2030. It estimates that data-centre owners would need $8.5 trillion in cumulative revenue through 2030 to cover announced capital spending, financing costs and an assumed 12% return on invested capital. Against that requirement, Gartner forecasts no more than $3.6 trillion in enterprise and consumer AI revenue attributable to AI data centres. The figures illustrate its projected return problem, but they are not an audited accounting shortfall: the 12% hurdle is an assumption, revenue may accrue to companies that do not own the underlying assets, and the public release does not provide enough methodological detail to reconcile every category.
Two of the numbers deserve a second look. The 12% return is a hurdle rate Gartner says it assumes; lower the hurdle and the required revenue falls with it. The $3.6 trillion is a forecast output, not a measurement — but the public release does not disclose the full model behind it (the release links readers to Gartner's client research), so the figure cannot be independently reconciled from the release alone.
Gartner has used different target returns in other AI infrastructure analyses, making the assumptions and scope of this particular 12% scenario especially important to understanding the result.
Two numbers, two windows
Gartner estimates that more than $3.5 trillion of AI infrastructure spending could be exposed to impairment by the end of 2030 if current investment trends continue. Its separate projection puts more than $4 trillion in asset value at risk of write-downs through 2032. These are exposure estimates over different periods, not cumulative losses that should be added together.
Who carries the risk
Gartner argues that hyperscalers' diversified revenue sources could help them sustain AI investment for longer, while specialized neoclouds face more immediate pressure to earn returns from AI-optimized infrastructure. It forecasts spending on AI-optimized IaaS rising from $66 billion in 2027 to $174 billion in 2030, yet predicts that more than 80% of neocloud providers will cease operating independently by 2030. That is a consolidation and survival forecast, not a prediction that more than 80% of providers will necessarily shut down. Lower capital spending, stronger revenue growth or different return assumptions could alter the outcome.
Until one of those moves, Gartner's case is less that the building stops and more that much of it ends up owned by someone other than the company that paid for it.
Sources
- [1] Gartner press release, 14 October 2026 — the $3.5 trillion exposure, the spending path, the $8.5 trillion requirement and assumed 12% hurdle, the $3.6 trillion revenue forecast, the $4 trillion through 2032, the IaaS estimates and the neocloud prediction; all figures in this piece are taken from this release and paraphrased, with no long quotationsRead source
- [2] The Register, “AI math isn't mathing: infrastructure spending forecast to outstrip expected revenue”, 14 October 2026 — secondary cross-check of the same figures, used only to confirm the primary release was read correctlyRead source