Gartner has put a hard number on a soft assumption: AI infrastructure spending looks set to outstrip the revenue it generates through 2030, leaving a gap of roughly $4.9 trillion by our calculation between what's needed and what's forecast. For UK buyers signing multi-year AI infrastructure deals, that's a reason to stress-test every payback assumption before the next invoice lands.
View the data behind this chart
| Revenue needed (12% ROIC) | Revenue forecast | |
|---|---|---|
| US$ trillion | $tn8.5 | $tn3.6 |
The scale of the shortfall Gartner has identified
Gartner's research puts AI infrastructure spend at nearly $1 trillion in 2026 alone, with a further $6 trillion forecast between 2027 and 2030. To hit a 12 percent return on invested capital — the bar many companies use internally to greenlight projects — Gartner calculates the industry would need to generate $8.5 trillion in AI-driven revenue over that period.
The problem is the revenue side doesn't come close. Gartner's enterprise and consumer AI revenue forecast for the same window tops out at $3.6 trillion — roughly 42 percent of what's required, by our calculation. That's not a rounding error; it's the kind of gap that forces write-downs, renegotiated contracts and, eventually, consolidation. Gartner's underlying report puts potential capital write-downs at more than $4 trillion by 2032 as newer GPU generations displace older hardware before it's been depreciated.
Neocloud GPU-as-a-service providers face the steepest cliff
The pain won't be evenly spread. Gartner expects hyperscalers to weather the shake-out reasonably well because AI is one revenue stream among several for them. Neocloud operators — the specialist "GPU-as-a-service" providers that exist largely to rent out AI compute — have no such cushion. Gartner forecasts that 80 percent of them won't survive as independent businesses beyond 2030, either folding or being absorbed by better-capitalised buyers.
Gartner senior director analyst Will Sommer was careful not to call it a blanket collapse: "Not everyone is in trouble. Some will outperform, some will underperform, and some will have to exit the market." For UK buyers who've committed to a neocloud for GPU capacity, that distinction matters — the question isn't whether the sector is in trouble, it's whether your specific supplier is one of the four-in-five, or the one-in-five.
The wider market is already getting nervous
This isn't an isolated warning. SoftBank found itself facing higher borrowing costs last month while trying to raise further funding for OpenAI, a signal that lenders are pricing in more risk around AI infrastructure debt. Financial analysts at Jefferies have flagged that model developers including OpenAI and Anthropic face their own margin pressure, from open-source alternatives and cheaper Chinese models undercutting them on price.
Management consultancy Bain & Company reached a similar conclusion from a different angle last month, estimating the AI industry needs $6 trillion in annual revenue by 2031 just to keep funding the infrastructure build-out, and warning of a shortfall unless genuinely new products and use cases emerge. Both the IMF and the Bank for International Settlements have separately raised concerns about an AI bubble and the knock-on effects a correction could have across the wider economy.
What this means for UK infrastructure buying decisions
None of this means AI infrastructure investment is wrong — it means the payback period assumptions baked into many 2026 procurement cases need revisiting. If your organisation is sizing a GPU cluster or signing a multi-year AI capacity contract, the Gartner numbers are a useful forcing function for asking harder questions about vendor durability, financing terms and realistic revenue attribution before committing capital.
Buyers evaluating NVIDIA DGX systems or other dedicated AI hardware should run the numbers using an AI GPU sizing calculator against actual workload demand, not vendor-supplied capacity targets, and cross-check the financial case with an IT finance calculator before signing. Reviewing AI server market data alongside your own utilisation forecasts helps separate genuine capacity needs from speculative over-provisioning.
A pre-commitment checklist for 2026 AI infrastructure spend
Given the scale of the forecast shortfall, due diligence on any AI infrastructure supplier — neocloud or otherwise — should go well beyond price per GPU-hour.
- •Check whether your supplier's revenue depends solely on AI compute rental, or spans a broader portfolio that can absorb a downturn
- •Model realistic AI revenue attribution for your own use cases, rather than assuming industry-wide growth projections apply to your deployment
- •Build exit and migration clauses into neocloud contracts in case a provider is acquired, restructured or exits the market before 2030
- •Factor GPU depreciation and refresh cycles into total cost of ownership, given the write-down risk Gartner has flagged as hardware ages
- •Revisit server configuration and GPU accelerator choices against actual, not forecast, workload utilisation