AI datacentres vs the grid: UK ministries clash over power needs — and the world is watching

How much electricity will artificial intelligence really gulp down? A fresh report has put two UK government departments on opposite sides of the socket. On April 26, 2026, the Guardian detailed a striking mismatch: the Department for Science, Innovation and Technology (DSIT) is planning for AI datacentres that could need around 6 gigawatts (GW) of capacity by 2030, while the Department for Energy Security and Net Zero (DESNZ) appears to foresee a fraction of that. In the same flurry of scrutiny, DSIT also revised its own estimate of AI-related emissions for the coming decade upward — dramatically. That’s not a rounding error; that’s a plot twist.

What actually happened

DSIT’s updated UK Compute Roadmap — the blueprint for building out national AI infrastructure — now explicitly forecasts “at least 6GW” of AI‑capable data‑centre capacity by 2030. For scale, that’s akin to adding multiple large power stations’ worth of demand in just a few years, with some sites envisioned at 500 megawatts or more and at least one zone scaling above 1GW. DESNZ, by comparison, has pointed to much lower sectoral growth figures in modeling cited in the Guardian’s report. Meanwhile, after questions from reporters, DSIT revised a document estimating the greenhouse‑gas footprint of AI compute to a range of roughly 34–123 million tonnes of CO₂ over ten years — a hundredfold swing from previous numbers. Translation: the government is now signaling that AI could be a non‑trivial piece of the national emissions pie.

Why this matters beyond Britain

AI datacentres are popping up everywhere, and the tug‑of‑war between climate targets and compute demand isn’t unique to the UK. In Europe, investigative reporting this month found that large US tech firms successfully lobbied Brussels to keep individual datacentre emissions data confidential — limiting public visibility into a sector set to triple capacity in five to seven years. If you’re trying to plan grids, attract clean‑energy investment, or simply trust that AI is being built responsibly, transparency matters.

The bigger global picture

Local power and planning strains are already visible. Councils in Sydney warned this month that clusters of datacentres could worsen blackouts, compete with housing near transit, and strain water supplies. It’s not that data halls are “the villain”; it’s that they bundle vast energy and cooling loads into tight geographic footprints. Combine that with AI’s spiking compute appetite, and cities that welcomed cloud campuses yesterday are asking tougher questions today.

Connecting the dots: from policy to plugs

Back in the UK, DSIT’s roadmap gestures toward a practical way through: purpose‑built AI Growth Zones that pair compute with dedicated clean power — think renewables, batteries, microgrids and, potentially, small modular reactors. Framed charitably, it’s an “if you’re going to do it, do it right” plan: co‑locate high‑density racks with reliable low‑carbon electrons, rather than bolt them onto already‑congested urban substations. That thinking is spreading globally as operators hunt for wind‑rich coastlines, solar‑heavy deserts, hydropower basins — even industrial‑park heat‑recovery schemes that warm nearby buildings. Expect more matchmaking between “where the watts are” and “where the GPUs want to be.”

What this means for everyday life

For consumers and businesses, this clash isn’t just about server farms — it’s about the AI features sneaking into your phone, your office apps, and your car. If power supply becomes the rate‑limiting step, cloud providers might nudge more inference (the day‑to‑day running of AI) closer to users to cut energy and latency, or throttle the splashiest features during tight grid windows. On the flip side, the scramble for clean power could accelerate grid upgrades, more rooftop and utility‑scale renewables, and new storage projects — the kind of boring, essential kit that quietly keeps your lights on and your video calls sharp. In other words: your toaster isn’t about to arm‑wrestle a datacentre for electricity — but it might benefit from a sturdier, greener grid built to feed AI’s appetite.

A quick, light reality check

Some breathless claims — “AI will eat the grid!” — are overcooked. Yet the UK’s own documents now plan for AI campuses measured in hundreds of megawatts each, while independent reporting shows industry pressure to limit emissions transparency in the EU. The prudent middle path is clear: build only where the clean power exists or can be added quickly, publish credible performance data, and reward efficiency from chip design to model architecture. Or, put more simply: fewer mystery numbers, more verified kilowatt‑hours.

What to watch next

• The UK’s forthcoming Carbon Budget 7 details this summer — do the energy and emissions math for AI finally align across departments?
• Announcements of specific “AI Growth Zones” and how they pair with renewables, batteries, or next‑gen nuclear.
• International moves on datacentre disclosure rules; if Europe reopens the transparency debate, other regions may follow.

Fresh angles to consider

Efficiency is a feature, not a footnote. Expect customers, regulators, and investors to reward providers that squeeze more performance per watt. New siting patterns — from energy‑rich rural hubs to ports with robust transmission — could reshape regional economies. And smarter demand timing (running big AI jobs when wind and solar peak) can make grids both greener and cheaper. The UK spat may look parochial, but it spotlights a global pivot: AI’s next breakthroughs won’t just be measured in model accuracy — they’ll be measured in megawatts, and who can supply them cleanly, transparently, and on time.