Gartner has put a number on the AI infrastructure boom: $2.4 trillion in global spend next year, up 53% on 2026, according to the firm's forecast. For UK buyers already fighting for GPUs, memory and grid connections, the forecast suggests buyers should not assume the supply squeeze will ease soon — see our AI GPU requirements calculator before committing budget.
View the data behind this chart
| 2026 | 2027 | |
|---|---|---|
| AI server spend | $527.6bn | $957.2bn |
Gartner's $2.4 trillion AI infrastructure forecast, decoded
Analyst firm Gartner has pegged worldwide AI infrastructure outlay at $2.4 trillion for 2027, sitting inside a broader $4 trillion figure for total AI expenditure that year, according to Gartner's forecast as reported by ITPro. Unveiled at the firm's IT Symposium/Xpo in Barcelona, the figure implies a 53% jump versus 2026 — faster growth than the 47.5% Gartner expects across AI spending as a whole. By this reckoning, infrastructure alone would account for nearly 60% of every AI dollar spent next year, per Gartner's estimates.
"Faster than anticipated expansion of user requirements into scaling up inference workloads, coupled with increasingly compute-intensive training workloads, continues to increase spending on AI infrastructure globally," John-David Lovelock, Distinguished VP Analyst and Chief of Research at Gartner, is quoted as saying, according to the cited Gartner/ITPro report.
Not every sub-sector is growing at the same rate. Domain-specific and specialised models are growing fastest, but general-purpose models still dominate the money.
- •AI cybersecurity spending: up 69% year-on-year, per Gartner's forecast
- •AI agents and assistants spending: up 123%, per Gartner's forecast
- •Generative AI model spending: up 82%, per Gartner's forecast
- •Domain-specific AI models: up 99% in 2027, per Gartner's forecast — yet foundation GenAI models still hold an 81% share of all model spend
Why this is a UK capacity problem, not just a budget line
For UK infrastructure buyers, the headline figure matters less than what it implies: every pound within that $2.4 trillion is competing for the same limited supply of chips, electricity and building space that domestic projects also depend on. The government has set a target of at least 6GW of AI-ready data-centre capacity by 2030, yet only around 2.9GW was estimated to be live in early 2026 — leaving roughly 3.1GW still to build just as hyperscalers and neoclouds everywhere ramp spending in line with Gartner's forecast.
The pipeline itself has swelled quickly: the number of committed UK schemes has climbed to 173, equating to more than 14GW, compared with 114 projects and roughly 10GW half a year earlier, according to Computer Weekly's reporting. Separately, around 4.4GW of projects are still awaiting planning approval, and more than 4GW of additional announced capacity is reported to be competing for grid connections — these categories are not necessarily additive, and electricity supply, rather than land or consent, appears to be the tighter bottleneck for much of the UK market.
It's worth checking real procurement numbers against this backdrop using our AI servers data study before scaling any deployment plan.
GPU and memory costs: the 'memflation' UK buyers are already paying for
Server hardware absorbs the brunt of Gartner's forecast. As the single biggest component of AI infrastructure outlay, spending on AI-optimised servers is set to almost double, climbing from $527.6 billion in 2026 to $957.2 billion in 2027. Lovelock put the surge down to hyperscalers and neocloud operators continuing to expand capacity, "as well as the ongoing 'memflation' of AI-optimised server prices."
That memflation isn't a theoretical concern. J.P. Morgan research cited alongside Gartner's numbers forecasts that DRAM pricing is projected to rise by more than 400% between the start of 2024 and the end of 2026, as manufacturers scramble to secure scarce memory supply. UK buyers weighing up enterprise GPU accelerators or NVIDIA DGX systems should model continued cost pressure into 2027 rather than assume prices will fall.
Scale matters too. One proposed UK AI and HPC facility tied to Nscale is reported to be sized for 50MW of dedicated AI/HPC load, with total power allocation reaching up to 90MW, and room for as many as 45,000 Nvidia GB200 GPUs, according to the project's announced specifications. A deployment of that size makes clear why any credible AI infrastructure budget must factor in power distribution, cooling and networking, not just the chips themselves — precisely the kind of stress test an AI GPU requirements calculator is built to run before procurement begins.

Where UK data-centre capacity is actually being built
The UK's current data-centre footprint sits disproportionately in London, reported to be home to roughly 40% of all facilities and close to two-thirds of estimated colocation capacity, according to market data cited by TechRadar Pro. That pattern is starting to shift, with Scotland and the North East among the regions landing sizeable new developments as developers chase power and sites beyond the London/M4 corridor.
Policy is adjusting in response. Under a draft UK national policy statement, data-centre projects of 100MW or above would fall under England's nationally significant infrastructure-project framework; the draft proposes a presumption in favour of approval, though this remains a consultation proposal rather than operative policy, with the consultation window closing on 8 January 2027. Beyond new construction, upgrading existing sites is proving a viable shortcut to faster go-live for operators stuck behind slow planning and grid-connection queues.
View the data behind this chart
| Layer | Detail |
|---|---|
| 2030 UK target | Government goal: at least 6GW of AI-capable data-centre capacity |
| Pre-planning pipeline | 4.4GW of additional projects not yet committed |
| Committed pipeline | 173 projects totalling more than 14GW, up from 114 projects/10GW |
| Live capacity | Around 2.9GW of UK data-centre capacity online in early 2026 |
What UK infrastructure buyers should do before committing capex
Put side by side, Gartner's forecast and the UK's own capacity figures suggest buyers should not assume the supply squeeze will ease soon: organisations that delay securing power, cooling and rack space until after ordering GPUs can leave themselves with equipment awaiting power, cooling or rack capacity. Getting the sequence right matters just as much as the size of the budget.
Practically, that means reserving grid capacity and physical infrastructure early, sizing accelerator needs against realistic workload profiles rather than peak ambition — our guide to provisioning for AI inference is a useful starting point — and weighing financing your AI build against outright capex given how fast server and memory prices are moving. Engaging IT procurement services early, and firming up server configuration decisions before placing orders, reduces the risk of over-committing capital while prices remain under pressure.
- 01ITPro — AI infrastructure spending to top $2 trillion next year · 16 November 2026
- 02Computer Weekly — UK committed datacentre pipeline grows to more than 14GW · 16 November 2026
- 03Computer Weekly — UK government drafts national policy statement for datacentres · 16 November 2026
- 04TechRadar Pro — UK data centre market lagging behind Europe · 16 November 2026
- 05DataCenterDynamics — How retrofitting can help win the AI infrastructure race · 16 November 2026
- 06DataCenterDynamics — UK AI Opportunities Action Plan data center · 16 November 2026
- 07computerweekly.com
- 08assets.theregister.com
