Capital expenditure on artificial intelligence infrastructure by technology firms is outpacing the returns that investment generates — a gap research firm Gartner Inc. projects at more than $3.5 trillion in potential write-offs before the decade ends, according to a report released Wednesday.

Gartner's analysis projects that AI providers are positioned to pour close to $1 trillion into hardware and infrastructure by 2026, with a further $6 trillion anticipated through the end of the decade, though the firm notes that revenue generation across that period may not be sufficient to cover those outlays. Gartner's own projection of roughly $3.6 trillion in AI-linked revenue across that period still falls below the threshold its senior director analyst Will Sommer identified as necessary for enterprises and consumers to recoup the capital expenditure currently flowing into AI infrastructure.

Gartner's analysis also projects that by 2030, more than 80% of neocloud operators — businesses that rent out specialised AI graphics processing units and high-performance cloud infrastructure — will no longer function as standalone entities, with the firm attributing that outcome to mounting cost pressures and an absence of sufficiently distinct business models relative to the revenues those providers can realistically generate. Neocloud companies draw their revenue almost entirely from AI-optimised infrastructure services, leaving them without the diversified income streams that give larger cloud providers a cushion against heavy hardware expenditure.

A Gartner forecast puts the share of independent neocloud providers likely to close or consolidate before 2030 at more than four in five, though the projection remains an estimate rather than a certainty. Projected spending on AI-optimised infrastructure-as-a-service is expected to reach around $66 billion by 2027 and approach $174 billion by 2030, though some analysts have noted that the speed of that growth may leave compute providers with limited time to recoup hardware investment before a given chip generation is superseded.

To build out its fleet of Nvidia GPUs and expand data center capacity, CoreWeave (CRWV) has drawn heavily on leveraged debt instruments and asset-backed financing structures rather than conventional capital channels. On Wednesday, the firm's unsecured bonds fell to pricing territory associated with financial strain. CRWV stock lost as much as 17 percent of its value at one point during Wednesday's trading.

Pressure on CoreWeave's debt instruments has not prevented the company from holding a market capitalisation north of $36.5 billion — a figure representing growth of more than 60 percent against its listing price when shares began trading in March 2025. CoreWeave chief executive Intrator told investors the current market behaviour reflects poor judgment while expressing continued faith in the AI cloud business.

Measured against the calendar year so far, CRWV has retreated 11 percentage points, IREN has shed 22 points, while NBIS has climbed 120 points.