The AI gold rush has a math problem, and Gartner just did the arithmetic.
The research and advisory firm says data-center owners need to generate $8.5 trillion in cumulative revenue by the end of the decade to justify what they are spending. Its forecast for the AI revenue those facilities will actually produce over the same stretch is no more than $3.6 trillion.
That leaves a $4.9 trillion hole.
The numbers behind the chasm
The warning comes from a Gartner report with a title that does not bury the lede: “Prepare for AI Market Consolidation: Capital Write-Downs Could Exceed $4 Trillion by 2032.”
On the spending side, Gartner projects AI providers will put nearly $1 trillion into infrastructure in 2026 alone. Another $6 trillion is expected to follow between 2027 and 2030, bringing the cumulative bill to approximately $7 trillion.
The $8.5 trillion revenue target is the figure Gartner says data-center owners would need to hit a 12% return on invested capital, or ROIC, by 2030.
Gartner’s revenue forecast falls far short of that bar. The firm expects total enterprise and consumer AI revenue attributable to these data centers to top out at $3.6 trillion by 2030.
Write-downs and a neocloud shakeout
Gartner warns that more than $3.5 trillion of AI infrastructure spending is at risk of impairment by 2030, a figure that could climb past $4 trillion by 2032.
Gartner analyst Will Sommer pointed to a structural issue: most AI revenue is being captured by asset-light companies, not the firms that own the infrastructure. As a result, Sommer expects returns on invested capital for infrastructure owners to stay below the 12% target until at least the end of 2032.
The report saves its starkest prediction for specialist AI compute providers, often called “neocloud” operators. Gartner predicts that more than 80% of neocloud operators may lose their independent status by 2030 through consolidation, acquisitions, or exiting the market entirely.
Diversified hyperscale cloud providers, by contrast, are expected to feel less financial strain, as their broad revenue streams give them a cushion that single-purpose AI compute shops do not have.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.