Opinion: Scotland Has Wind Power. AI Data Centres Still Need a Grid

The AI industry has a habit of presenting a data centre as a financial commitment. Announce the site, name a technology partner, count the chips and let the power supply sound like an engineering detail. Microsoft’s reported exit from Nscale’s planned Loughton campus shows why that order of thinking is backwards. An AI facility is real only when it can be connected, powered and used.

Scotland offers a more interesting test of that proposition than another generic warning about electricity demand. It has substantial renewable-energy potential and an explicit ambition to attract green computing. But Scotland’s 2026-2031 AI strategy acknowledges that generating renewable power and delivering it to a particular data centre at the right time are different things. Its government says the grid and access to it must change if major investments are to materialise.

That is the paradox Scotland should put at the centre of its pitch. A windy country does not automatically provide a ready socket for a hyperscale campus. Transmission, substations, connection queues and the timing of new capacity still determine whether a proposed site can operate. Renewable abundance is an advantage, but it is not a substitute for a credible connection date.

The Scottish Government has already moved to examine large projects more closely. A direction effective from September 17 requires an environmental impact assessment for new data-centre proposals exceeding 50 megawatts of power capacity. Smaller developments remain subject to case-by-case consideration. That is a meaningful step because it brings energy, water and hardware impacts into the planning conversation before promises harden into political commitments.

But environmental assessment should be the start, not the entire test. I would require each hyperscale proposal seeking public support to publish a realistic power-delivery schedule, identify who pays for associated network upgrades, and explain how it would operate if the connection is late. Any claim of green computing should disclose whether the site will actually use low-carbon electricity at the relevant times, rather than simply point to renewable generation somewhere on the grid.

Scotland’s strategy also looks beyond giant campuses. It argues that distributed computing could be deployed more quickly in suitable places and could create opportunities for local AI use and heat reuse. That deserves attention. A portfolio of smaller, well-sited facilities will not replace every frontier-training cluster, but it may bring some capacity closer to researchers and businesses without making every project depend on a single enormous grid connection.

This is not an argument to stop building AI infrastructure. It is an argument to stop measuring progress by announcements alone. The useful metric is delivered computing capacity that works for customers without shifting hidden power or environmental costs onto communities. A deal with a famous cloud company is not proof of that.

The distinction matters in African markets too. TechBooky has argued that communities need to see the real benefits and costs of data centres. Where grids are weaker, the temptation to build private power islands is even greater. Scottish policy offers a different question to ask early: can the network, the planning system and the host community support this project on the timeline being advertised?

The reported Loughton setback is an English story, not evidence that a Scottish project has failed. But it should make Scotland more disciplined about the projects it welcomes. The region that wins the next phase of AI will not necessarily be the one with the loudest compute announcements. It may be the one that can connect useful capacity, document the costs and keep its promises.