Scotland's green datacentre push faces AI emissions test

Scotland’s attempt to attract green datacentres is facing a harder AI-era test after Action to Protect Rural Scotland said the framework behind the label predates ChatGPT and leaves the emissions cost of hyperscale computing outside the climate case used to justify it.

The dispute reaches beyond Edinburgh. Scotland wants AI investment, and its AI strategy treats green data centres as part of the infrastructure needed to support that goal. But the economics have shifted faster than the planning language. The International Energy Agency says global data-centre electricity use could reach 945 TWh by 2030 in its base case, while Reuters reported that total U.S. power demand is projected at 4,195 billion kWh in 2025, 4,244 billion kWh in 2026 and 4,381 billion kWh in 2027 as AI use rises. A planning label that looked narrow in 2022 now sits in the middle of a broader argument about power.

Ministers argue that Scotland is well placed. A spokesperson said Scotland has “significant strengths as a location for green datacentres” because it has abundant renewable energy, a skilled workforce and resilient fibre links, according to The Guardian’s reporting. Investors looking at AI capacity tend to want power, land, cooler weather and political backing. Scotland can plausibly offer those things.

Still, the official definition was built for a smaller industry. In National Planning Framework 4, the Scottish government treats green datacentres as developments with an overall negligible effect when set against their wider economic value and potential use of renewable power. Campaigners say the analysis behind that position was done in 2022, before ChatGPT turned AI infrastructure from a niche technology issue into a power-planning issue. In practice, the term green was fixed in policy before ministers had to reckon with what always-on AI workloads might demand from the grid.

To APRS, that gap is the issue. The group is not arguing that data centres should never be built. It says the current climate accounting is too permissive for an industry now being expanded in the name of AI. Planning advantages are not granted to an abstract technology. They are granted to large physical sites with heavy electricity loads, backup-power needs and local land-use effects.

Kat Jones, director of Action to Protect Rural Scotland, told The Guardian that “it is pretty shocking to find out that the vast carbon footprint of hyperscale datacentres has been completely excluded from the greenhouse gas analysis for our planning framework.”

The argument carries weight because it is aimed at process rather than rhetoric. APRS is pressing for a moratorium until ministers examine the effect on climate targets, local communities and the electricity system. The same Guardian analysis cited by campaigners says more than 100 UK datacentre projects have requested gas connections. If even part of that buildout relies on fossil-backed power at the margin, the green label starts to look less like a measured standard and more like an incentive left over from an earlier phase of the market.

A policy written before ChatGPT

Across governments, this is becoming a familiar pattern. Officials speak about AI infrastructure in the language of growth policy, strategic capacity and digital sovereignty, while environmental tests are still being applied through older planning categories. Scotland’s own documents show the same gap. The country’s 2026-2031 AI strategy treats green data centres as enabling infrastructure and says projects should secure renewable power and grid access. It does not appear to publish a simple public standard for when a data centre is green in operational terms and when it is merely well situated.

At Holyrood, Ariane Burgess has tried to force that question into the open. The Scottish Greens MSP for Highlands and Islands is less focused on whether AI should exist than on whether ministers are using a flattering category without spelling out the trade-offs. Any credible definition would have to say more about load, backup fuel, the timing of renewable supply and the pressure on transmission upgrades.

Burgess told The Guardian that “we urgently need transparency around what constitutes a ‘green datacentre’ and how their huge energy demands will be accommodated by our grid infrastructure.”

In infrastructure policy, labels do real work. They can shape permitting, make projects easier to defend politically and soften the conflict between industrial policy and climate targets. If a government wants the economic upside of AI data centres, the pressure is to describe them as aligned with the energy transition. If critics want stricter tests, the pressure is to move the argument from branding to accounting.

A version of that competition is visible elsewhere. CNBC argued this week that Gulf states have pitched themselves as AI hubs by leaning on cheap power and strategic geography. Scotland’s version of that pitch is cleaner and colder: renewables instead of hydrocarbons, grid access instead of giant domestic fuel reserves. The contest underneath it is similar. Places that can promise electricity are trying to turn that promise into AI investment before the planning costs are fully fought through.

The grid question behind the label

The Scotland row also shows how quickly the centre of gravity has moved. For years, debates about data centres often turned on tax breaks, jobs and broadband capacity. AI has pulled the argument toward energy first. Once ministers and investors start talking about hyperscale facilities, the central question is no longer whether a country is digitally ambitious. It is whether it can add large, continuous demand without shifting costs, delays or emissions somewhere else in the system.

Those wider energy numbers do not decide the Scottish case on their own, but they give campaigners a stronger backdrop than they had a few years ago. The IEA’s estimate of 945 TWh by 2030 makes clear that AI is no longer an abstract future-demand story. In the United States, Reuters’ reporting on EIA forecasts tied record power demand directly to data centres serving AI and cryptocurrency.

Other resource constraints are entering the same debate. In Chile, The Guardian reported that the datacentre boom is worsening water stress in areas already hit by drought. Scotland faces a different pressure point, but the policy question is similar: what local resource is actually scarce once AI infrastructure arrives at scale?

Seen that way, Scotland is an early test case. It is not the biggest market, and it is not building on the scale of the United States or the Gulf. Even so, it shows an argument many governments are trying to postpone. A country can advertise renewable power, cooler temperatures and pro-growth planning. It still has to say what happens when AI demand starts arriving faster than the old carbon assumptions.

For now, ministers are defending a framework designed to attract investment, while campaigners are trying to force a more literal reading of the word green. Both sides are arguing about timing. The government’s case assumes the planning label can stand until the projects arrive. Critics say the accounting will be too late by then.

That leaves Scotland with a policy problem likely to spread beyond its borders. If a government wants to market datacentres as green in 2026, it will need more than a favourable location and a clean-energy story. It will need a test that survives AI-era power demand, and a public explanation of who pays when that test gets harder.