The UK wants to be an AI powerhouse, with Chancellor Rachel Reeves even pledging that it will have the fastest AI adoption out of any G7 country. But if we want to actually have ambition meet reality, in other words, have more AI adoption, we need more compute. And if we want more compute, we need more data centres. And if we want more data centres, we need more land, more cooling and, above all, a lot more power. 

That is why Carbon Brief’s analysis this week hit such a nerve. It focused on a point that should already have raised eyebrows in Whitehall – the Government’s own numbers on data centre emissions looked far too low. And on that narrow but important point, Carbon Brief is right. 

The original DSIT Compute Evidence Annex said UK AI compute demand could reach 11.2GW by 2035 while associated emissions would still come in at just 0.025 to 0.142 MtCO2. That was laughably inaccurate, so much so the Government has already quietly updated the forecast. Not with a more accurate forecast, but with a note stating that it was intended to inform policy development rather than represent a final cross-government view, and that it’s now working on a more accurate assessment.

The Government’s numbers never really added up

You don’t need to be anti-data centre to see the problem. The basic logic is obvious enough. Huge amounts of AI compute require huge amounts of electricity, and electricity is only as clean as the system supplying it. The Government is right to talk about AI Growth Zones and faster infrastructure delivery, but it clearly got ahead of itself when it implied the emissions side of the equation would be negligible. If Britain is serious about scaling compute, then the power and carbon implications have to be taken seriously too.

It is not hard to see why the modelling ended up looking so convenient. The Government has staked a lot on AI as a driver of productivity and growth. It’s also made bold claims when it comes to its commitment to driving down carbon emissions and ensuring a greener grid. It is possible that officials simply assumed grid decarbonisation would move fast enough to absorb the coming wave of data centre demand. Maybe it still will. But that’s an assumption that still has to be proved, not wished into existence, and if we’re going to do forecasts – they should probably be based on probability. 

Ofgem’s own figures show just how big of a challenge it would be to get anywhere close to the UK Government’s previous forecasts. The regulator says there are around 140 data centres in the queue representing roughly 50GW of demand, including 71 projects amounting to around 20GW that have already reached financial commitment with final investment decision. Of course, not all of those data centres will come to fruition, but there are new projects proposed on a near daily basis – so the important thing to remember is that we’re getting a large new fleet of data centres regardless. 

Carbon Brief is right – but not about everything

I don’t want to get bogged down on Carbon Brief’s analysis being right or wrong, because it’s useful regardless, even if not flawless. That’s because it’s right to point out that the Government’s numbers looked too rosy, but it’s probably being a bit hyperbolic when it suggests that Britain is heading for a dirty, gas-fuelled data centre boom. In fact, just like the Government’s forecast, its ‘hundreds of times higher’ headline is based on a scenario, not a forecast. More specifically, it relies on cases where a meaningful share of future data centre electricity comes from gas. That is a perfectly fair stress test. It is not the same thing as saying this is the most likely outcome. Again, probability matters. 

There is also an important like-for-like problem in the comparison. Carbon Brief compares DSIT’s 11.2GW AI-only figure with Ofgem’s broader estimate that 71 mature data centre projects amount to around 20GW in the connections pipeline. Carbon Brief itself acknowledges that these figures are ‘not directly comparable’, because the Ofgem number is not specifically AI-only. That caveat matters. If the question is whether the Government’s modelling looked too low, the answer is yes. If the question is whether the sector is therefore heading for a fossil-fuelled free-for-all, the evidence is far less clear.

There is another point worth making here too. Backup generators are a real environmental issue, but they are not the same thing as routine power supply. Parliament’s latest POSTnote notes that generators are typically emergency systems, used infrequently and tested monthly, with regulatory limits on non-emergency run hours for larger installations. So it is important not to collapse every discussion of backup fuel into a claim that data centres are routinely running on-site fossil generation as their main source of electricity. That is not how most of the sector operates.

This is really a grid story

That is why the real story here is not whether data centres are ‘good’ or ‘bad’. It is whether Britain can build enough clean power, network capacity and system flexibility in the right places, quickly enough, to support the next wave of digital infrastructure. Before you argue that I’m simply defending data centres due to the fact that I write for a data centre publication, I can assure you I’m not. The industry can only get better if it’s held to account, but that doesn’t mean we can just keep adding fuel to the fire that simply states data centres are automatically bad. 

The public debate still swings too easily between two lazy positions. One says data centres are much cleaner than other forms of heavy industry and therefore should be given a pass. The other says they are climate villains in waiting. Neither is accurate. 

As Arcadis’ David Field argued recently, it is time to separate fact from fiction on data centre energy demand, because the sector only really makes sense when viewed in its wider system context. Data centres are energy-intensive, yes, but the real question is how grid capacity, phasing, cooling design and long-term energy planning evolve around them, not whether a single headline number can do all the work.

In fact, data centres will only be as low-carbon as the energy system, planning regime and technology choices around them allow them to be. That is why AI Growth Zones matter. The Government’s own response to the AI Opportunities Action Plan says these zones are supposed to offer enhanced access to power and support for planning approvals, while taking energy requirements into account with the National Energy System Operator. The AI Energy Council was also set up specifically to look at how AI and clean-energy goals can be delivered together.

The industry is already moving

And this is where the industry side of the story deserves more airtime than it usually gets. It’s not like the UK data centre sector is sitting still waiting to be told it has an emissions problem. It’s already well aware and is moving quickly on procurement, energy efficiency and backup power. 

Equinix, one of the biggest players in the space globally, says its UK data centres have 100% renewable energy coverage on a market basis, that all new UK sites since 2021 no longer use natural gas for heating, and that its Manchester 5 facility has used HVO for backup generators since 2022. It is also trialling lower-GWP refrigerants and says it is maintaining a focus on further PPAs and on-site low-carbon energy options.

Equinix is not alone. Kao Data says it procures 100% renewable energy on a market basis, reports an average estate PUE of 1.53, says it pioneered HVO for backup power in the UK, and has switched its Renewable Energy Guarantees of Origin (REGO)-backed supply to Dogger Bank from April 2025. VIRTUS says it has used purely renewable energy since 2012. Those are not magic fixes, but they do show the sector is actively working the levers it can control.

This is also where it is worth being honest – after all, I promised to hold the industry to account. Procuring 100% renewable energy on a market basis is not the same thing as saying a facility is physically running on zero-carbon electricity every hour of the day. Critics are right about that. Equinix’s own UK reporting shows market-based Scope 2 emissions at zero, while location-based Scope 2 emissions remain material. Kao Data reports the same distinction, saying its market-based Scope 2 emissions remain at zero because of REGOs while location-based emissions remain significant and may rise as the business grows. So yes, the grid still matters enormously. That is why it’s not an argument against data centres. It is an argument for getting the wider energy system right.

A plan is starting to emerge

There are also signs that the industry is trying to go beyond certificates and easy fixes. Kao Data’s deal with Downing Renewable Developments to build a 40MW solar farm for its Harlow campus is a good example. The project is designed to supply the campus with solar-generated electricity via a private-wire arrangement under a long-term PPA. It is not a whole-sector solution on its own, but it does point to a pattern of larger operators taking things into their own hands by being more direct with their renewable energy procurement and reducing the pressure on the grid, rather than simply complaining about it. 

We’re also seeing similar agreements from other operators, with SMRs seen by some in the industry as an ability for the sector to decouple from its reliance on the grid, while also reducing carbon emissions. While we’re still some way from seeing the first SMR deployed in the UK, Holtec International, EDF UK and Tritax Management have agreed to develop an SMR in Cottam, Nottinghamshire, for the express purpose of powering a data centre. 

There is also a broader framework taking shape. techUK says the current Climate Change Agreement for the sector requires a 14.5% energy-improvement target by 2030 against a 2022 baseline. The Climate Neutral Data Centre Pact, which many operators have signed up to, says electricity demand should be matched by 75% renewable or hourly carbon-free energy by the end of 2025 and 100% by the end of 2030. Again, these are not slogans. They are formal benchmarks against which the sector can be judged.

And if policymakers want examples from abroad, Ireland is already moving in a more explicit direction. Its regulator has decided that new data centres connecting to the electricity network must provide generation and/or storage capacity, and must meet at least 80% of their annual demand with additional renewable electricity projects generated in the Republic of Ireland. Given the scale of the power problems in Ireland, it could serve as a testbed for how to deal with a growing need for AI data centres without breaking the grid. 

The real question 

So yes, the Government probably did underestimate data centre emissions. Carbon Brief was right to say so. But it is too simplistic to jump from that to the conclusion that the UK is heading for a reckless, fossil-fuelled data-centre boom. The more accurate picture is messier, but also more constructive.

The official numbers need fixing. The grid needs to move faster. Siting decisions need to get smarter. And the industry needs to keep proving that its decarbonisation plans are real, measurable and not just market-based spin. But the outline of a plan is already there: cleaner procurement, lower-PUE design, cleaner backup fuels, more direct renewable deals, tougher sector targets and a bigger push to build where the power system can actually support growth. 

The real failure now would not be a lack of ambition from the industry. It would be a failure from the Government to match that ambition with an energy strategy capable of making it credible.