Whitehall Splits 10x on AI Datacentre Power, Bleeds UK Net Zero

The procurement lead at a Manchester managed-services firm called me on Friday with one question: "Whose number do I use?" She had two pricing-model spreadsheets open. One assumed UK datacentre demand growing roughly with the wider commercial-services sector. The other assumed an AI-specific buildout an order of magnitude larger. Same firm, same client, same year. Two answers that don't fit in the same room.

That's the mood the Guardian's reporting just made official. The Department of Science, Innovation and Technology says the UK will need at least 6GW of AI-capable datacentre capacity by 2030. The Department of Energy Security and Net Zero, when asked how it had baked AI datacentres into the country's emissions trajectory, pointed to a forecast for the entire commercial-services sector growing 528MW over the same window. A single one of DSIT's planned AI growth zones is sized at 500MW.

The numbers don't reconcile. They're not close.

The Deployment

What's actually on the table, per the Guardian and the underlying departmental documents, is this. DSIT, in its 2025 UK compute roadmap, committed to "at least 6GW" of AI-capable datacentre capacity by 2030, sourced from multiple AI growth zones across the country, each requiring at least 500MW of electricity. DESNZ, which owns the carbon budget growth and delivery plan, was asked by the NGO Foxglove how it had incorporated AI datacentres into its emissions projections. The department's response: it doesn't hold separate projections for datacentre growth, and pointed to a commercial-services-sector forecast of 528MW additional load between 2025 and 2030.

The Guardian flags the obvious problem. DSIT's headline number is roughly ten times the figure DESNZ is using to plan the country's path to net zero. One AI growth zone, on its own, would consume nearly the entire commercial-services-sector load increase that DESNZ has modelled.

There's a second deployment, quieter but sharper. DSIT's original compute-roadmap annexe projected 10-year cumulative emissions from additional AI computing capacity at 0.025 to 0.142 MtCO2, under 0.05% of UK projected emissions. That document was pulled from the government site earlier this year after Carbon Brief queried the math. After the Guardian asked about it last week, DSIT posted updated figures: 34 to 123 MtCO2 over ten years, or 0.9 to 3.4% of UK projected emissions. That is not a rounding revision. That is a more-than-hundredfold upward correction, published the day after press contact.

Tim Squirrell at Foxglove called the gap between departments "cluelessness" that "would be laughable, if it weren't so alarming." Cecilia Rikap at UCL offered the sharper read: either the two departments are incompetent, or there's "magical thinking about AI and big tech."

Why It Matters

The vendor pattern this echoes most directly is the German grid-connection queue from the early-2020s renewables build, when economy and environment ministries published interconnect forecasts that didn't agree and the operator class, industrial customers, utilities, colocation buyers, discovered the gap by getting connection denials they'd assumed were already budgeted for. The lesson was the same then as now. When two ministries disagree on a load forecast by an order of magnitude, the people who feel it first are not the ministries. It's the operator trying to sign a power contract against a tariff that one of those forecasts implicitly underwrites.

For UK SMBs and mid-market firms, the question isn't whether the 6GW gets built. It's which department's number ends up sitting underneath the carbon budget, the grid-reinforcement plan, the planning permissions for the growth zones themselves, and, eventually, the wholesale electricity price they'll pay for the next decade. DESNZ's carbon budget 7 is due this summer. A DESNZ spokesperson told the Guardian that "datacentre emissions are factored into our modelling, including for carbon budget 7." That sentence is doing a lot of work, given the department's own response to Foxglove was that it doesn't hold separate projections for datacentre growth.

There's a buyer-landscape read here too. The hyperscalers and the Mittelstand-equivalent UK colo operators that will fill these growth zones, the Equinix, Digital Realty, Ark, Kao classes of buyer, already have a clearer picture of their own pipeline than the carbon planner does. That asymmetry is the actual story. The corporations know what they're building. The department charged with fitting it into a national emissions plan is using a sector-aggregate forecast that AI datacentres are quietly making obsolete in real time.

What Other Businesses Can Learn

If you operate a UK SMB that buys compute, signs colo contracts, or is sitting near one of the announced AI growth zones, three things are now actionable.

First, treat your power-cost forecasting horizon as a moving floor, not a planning assumption. The honest read of the Guardian's reporting is that the carbon-budget arithmetic that flows into wholesale electricity pricing for the back half of this decade is openly contested at department level. If your three-year operating budget assumes UK industrial electricity pricing tracks the current trajectory, model a sensitivity case where it doesn't. The 6GW number, if it lands, has to be paid for by someone, and that someone is usually the customer downstream of the grid reinforcement.

Second, on colo and hosting contracts being signed in 2026, read the green-energy clauses with a sharper eye than you did last year. Several UK colo providers are pricing in 100%-renewable power as a contractual headline, but the underlying grid mix and the offset arithmetic depend on whose forecast the regulator and Ofgem are actually using. If your provider's sustainability commitments are sourced from the lower DESNZ number, you have a paper-trail risk that gets worse as DSIT's 100x emissions revision works through the system.

Third, for any operator considering siting near an AI growth zone for the assumed grid reliability or capacity headroom, verify the connection date independently. The growth-zone announcement is a policy document. The actual interconnect is a National Grid problem. Those two timelines have a long history of not agreeing, and a 6GW vs 500MW gap between departments is exactly the kind of thing that turns into a delayed connection two years from now.

When two ministries disagree on a load forecast by an order of magnitude, the people who feel it first are not the ministries.

The cleanest move for a small operator right now is to assume DSIT's number is closer to the build-out reality and DESNZ's number is closer to the planning reality, and to budget the gap between them as a risk premium on every UK power-linked contract you sign this year.

Looking Ahead

Carbon budget 7 lands this summer. That document is the first place we'll see whether DESNZ has actually re-modelled with DSIT's 6GW number, or whether the gap survives into a published plan. Watch the AI Energy Council's output between now and then; that's the body the DESNZ spokesperson named as "exploring opportunities to attract investment and support the development of clean power for datacentres."

I asked the Manchester procurement lead what she'd told her client. She said she sent both spreadsheets and a one-line note: "Pick the higher one and assume we'll explain why later." Then she went back to reading the compute roadmap, the version that's still on the government website.

Sources

• UK departments at odds over energy demands of AI datacentres, accessed 2026-04-27