Gartner: $3.5 Trillion of AI Infrastructure Spend Faces Impairment by 2030

On 14 October 2026 Gartner put a number on the AI revenue gap. It said more than $3.5 trillion of AI infrastructure spending could be exposed to impairment by the end of 2030 if investment continues at its current pace, according to its press release and Network World's report on it. By Gartner's arithmetic, AI data center owners need about $8.5 trillion of cumulative revenue through 2030 to justify what they plan to spend. It forecasts that enterprise and consumer AI revenue attributable to those data centers will reach only about $3.6 trillion. Gartner also predicted that more than 80% of neocloud GPU providers will stop operating independently by 2030. The release came out on the same day that CoreWeave shares fell 15%.

What happened

The analysis comes from Will Sommer, a senior director analyst at Gartner. Its starting point is the spending plan. AI providers are expected to spend nearly $1 trillion on AI infrastructure in 2026 and about $6 trillion more between 2027 and 2030. To cover that capex, the financing costs and the expected returns, data center owners would need roughly $8.5 trillion in revenue by 2030, CIO.com reported. At current growth rates, Gartner sees enterprise and consumer AI application revenue reaching $3.6 trillion. That leaves a gap of nearly $5 trillion.

"Enterprise and consumer AI revenue will be insufficient to support profitable returns," Sommer said. Gartner's view is that the gap mostly closes through less spending, as the neocloud market consolidates and investment falls back in line with demand. It expects more than $4 trillion of asset value to be exposed to write-downs through 2032. It does not expect the industry's assumed 12% return on invested capital to be reached before 2032 at the earliest.

Neoclouds come off worst in the forecast. Gartner defines them as companies that build or run AI-optimised infrastructure and rent out accelerators and high-performance compute, often in competition with hyperscalers. Mounting costs and difficulty differentiating will push more than 80% of them out of independent operation by 2030. Network World named CoreWeave, Nscale, Nebius and Lambda as examples of the category. Gartner expects the AI-optimised IaaS market itself to grow from $66 billion in 2027 to $174 billion in 2030.

The numbers

Why it matters

Most bubble arguments compare capex with revenue. This one goes further and asks whether the assets will be worth their book value when the bill comes due. That turns on how long a GPU stays useful. In early October, Reuters reported (as republished by CTech) that banks usually underwrite GPUs over three to four years. Major cloud companies are reportedly stretching server depreciation to five or six years, and Nvidia says its top GPUs can earn revenue for about a decade. If the lenders turn out to be right, the write-downs Gartner describes arrive sooner.

The timing was awkward. CoreWeave fell 15% that day and its credit default swaps passed 1,000 basis points, Reuters reported, although Reuters tied the move to inflation data, rate expectations and doubts about AI profits rather than to Gartner. Reuters later reported that CoreWeave, Nebius and IREN each fell more than 30% in October.

What sceptics say

Gartner hedges its own forecast. "The scenario described in our research is the most likely scenario given current trends, but not the only possible scenario," Sommer said. The revenue side of the forecast is also the hardest part to predict. In August Nvidia guided to $108.0 billion of revenue for a single quarter, per its release, and CoreWeave said in September it was signing new capacity at higher prices. A revenue forecast that runs to 2030 is the input most likely to move as real adoption data comes in. And an impairment exposure is a ceiling on what could be marked down, not a forecast that it will be.

My take, as opinion: "80% of neoclouds stop operating independently" mostly means mergers and acquisitions, not data centers going dark. The GPUs will still run. What changes is who you sign the contract with, and on what terms.

What it means for engineering and DevOps teams

• Plan for your GPU provider changing hands. If a forecast like this is even half right, your provider may be acquired during your contract. Check assignment and change-of-control clauses, and keep your isolation and runtime choices portable.
• Expect cheaper compute late in the decade, not now. An overbuild usually ends with discounted capacity, but the timing is the uncertain part. Keep commitments short enough to benefit if prices fall.
• Measure value per GPU-hour. The gap Gartner describes is ultimately about whether AI features pay for themselves. Teams that can show cost per useful output, as in FinOps for AI, will keep their budgets when the CFO starts asking.

FAQ

What does Gartner's $3.5 trillion impairment figure mean?

It is the amount of AI infrastructure spending Gartner thinks could be exposed to asset write-downs by the end of 2030 if spending continues at its current pace. It is an exposure estimate, not a forecast that $3.5 trillion will be written off.

What is a neocloud, and why does Gartner expect most to disappear?

A neocloud is a provider that builds or operates AI-optimised infrastructure and rents out accelerators, CoreWeave being the best-known example. Gartner expects cost pressure and weak differentiation against hyperscalers to push more than 80% of them out of independent operation by 2030, mostly through consolidation.

Is the AI revenue gap real?

Gartner's estimate is a gap of nearly $5 trillion through 2030, between about $8.5 trillion of revenue needed and $3.6 trillion forecast. Other forecasters disagree on the revenue side, and spending plans can be cut. That is how Gartner itself expects most of the gap to close.

Related: GPU Clouds in 2026: H100s at $2.59 an Hour, $104B of CoreWeave Backlog on $35B of Debt, and a Six-Year Bet on Silicon, AI Power in 2026: 485 TWh, a 0.3 GW Stargate, a 116 GW Turbine Backlog, and 1,800 MW That Dropped Off the Grid in Seconds and Self-Hosting LLMs vs API Costs in 2026: The 62% Utilisation Break-Even.