Gartner: $3.5T AI Write-Down Risk by 2030

Gartner says more than $3.5 trillion of AI infrastructure spending could be exposed to impairment by the end of 2030 if AI providers keep spending at their current pace. An impairment is a write-down: the company admits an asset is worth less than it paid and takes the loss on its books. The forecast comes from a Gartner press release published on October 14, 2026.

The core of the warning is a revenue gap. Gartner calculates that AI data center owners need $8.5 trillion in cumulative revenue through 2030 to justify what they have announced, but forecasts that AI revenue attributable to those data centers will reach no more than $3.6 trillion.

Gartner’s forecast in numbers
Measure	Gartner figure	What it covers
AI infrastructure spending	Nearly $1 trillion in 2026, plus $6 trillion from 2027 to 2030	Capital spending by AI providers
Revenue needed	$8.5 trillion through 2030	Covers announced spending, financing costs and a 12% return on invested capital
Revenue forecast	No more than $3.6 trillion	Enterprise and consumer AI revenue attributable to AI data centers
Write-down risk	More than $3.5 trillion by 2030, more than $4 trillion by 2032	Assets that may have to be written down if spending continues on its current path
Neocloud outlook	Over 80% cease independent operations by 2030	Specialist GPU cloud providers

Table of Contents

How Gartner arrives at the gap

Gartner’s $8.5 trillion figure assumes data center owners want a 12% return on invested capital (ROIC), which it treats as the minimum return investors expect. On top of building costs, that number includes interest and financing on the money borrowed to build.

The revenue side is smaller partly because of who earns the money. Many AI companies rent computing power rather than own data centers, so Gartner predicts less than half of AI revenue will go to providers that own significant AI infrastructure.

In their words

“Enterprise and consumer AI revenue will be insufficient to support profitable returns on all planned AI capital expenditures during this period.”

Will Sommer, Senior Director Analyst at Gartner, in the Gartner press release

In this scenario, Gartner does not expect the industry to reach its 12% return until the end of 2032 at the earliest. Gartner is not the only firm making this case: Network World notes that Bain recently estimated AI infrastructure spending could reach $1.5 trillion a year by 2031, with AI services needing to top $6 trillion a year to pay for it.

Why neoclouds are most exposed

Neoclouds are cloud providers built mainly around renting out GPU computing for AI, as opposed to hyperscalers such as Microsoft Azure, AWS and Google Cloud that sell a broad range of services. Fierce Network lists CoreWeave, Nebius, Lambda and Nscale among the smaller providers racing to add capacity.

Hyperscalers

Diverse revenue buys time

Gartner says hyperscalers have large, varied revenue streams that could plausibly sustain AI spending for years.

Cushion: Losses on AI capacity can be absorbed by other businesses.
Timeline: They can wait longer for AI revenue to catch up.

Neoclouds

Returns are needed soon

Neoclouds depend on renting AI compute, so they must earn a return on it quickly.

Market size: Gartner sees AI-optimized infrastructure-as-a-service spending growing from $66 billion in 2027 to $174 billion in 2030.
Hardware clock: Gartner says that growth is too slow for most providers to recover their costs before today’s AI chips are replaced.

Gartner predicts cost pressure and a lack of a distinctive business model will push over 80% of neoclouds to stop operating independently by 2030, which in practice means being acquired, merging or closing. Sommer told Fierce Network that the chips inside these data centers typically last five to six years, so a costly refresh could land around the time revenue starts to catch up.

Recent news shows how exposed single projects can be, even for reasons other than revenue. On October 18, the Guardian reported that Microsoft had ended its agreement for Nscale’s planned AI data center in Loughton, Essex, after the site was told grid power may not arrive until the 2030s.

What it means for IT teams

For organizations renting GPU capacity, the practical risk is provider concentration. If a smaller provider is acquired or shuts down, workloads may need to move on short notice, so it is worth checking contract terms on exit, data export and notice periods before signing long commitments.

For Azure customers, Gartner’s view is that hyperscalers are better placed to keep investing, though it does not single out any provider. Gartner clients can read the full research, titled Prepare for AI Market Consolidation: Capital Write-Downs Could Exceed $4 Trillion by 2032, and the firm says it will present more at its IT Infrastructure, Operations and Cloud Strategies conferences in London (November 16 to 17), Tokyo (December 1 to 3) and Las Vegas (December 8 to 10).