What does AI mean for Wales?

Last week I went along to the launch of the UK’s £500m Sovereign AI fund (sovereignai.gov.uk), and I came away genuinely impressed. Credit to UK Government for standing up what is essentially a fast-moving VC arm this quickly. I like that it’s built to help innovative AI startups get over the hurdles they face selling into the public sector, and the offer that really catches the eye is up to £2m of GPU compute for the companies it backs, delivered through the new Isambard-AI data centre in Bristol. So the question I keep coming back to is: what does this mean for Wales?

My first answer is that we need a ‘Team Wales’ approach if we’re going to make the most of it. Leave our entrepreneurs and startups to make unsupported applications to Sovereign AI on their own and I’m confident we’ll see very little of that investment land here. We have to create the conditions for success, and to my mind that means five things.

• A real research base building the foundation models for the sectors where we can genuinely compete; ie medical discovery, energy, fintech, mediatech, Compound Semi Conductors and the like. That means our best researchers working alongside our strongest businesses, pulling in UK and EU grant money and pairing it with VC funding of the sort that Sovereign AI provides.

• Anchor AI lighthouses. These are the world-renowned businesses that catalyse a whole sector. Admiral did exactly this in Cardiff, it became the beacon that spawned a cluster of insure-tech startups and a thriving ecosystem around them. Worth noting that one of Sovereign AI’s first investments is in Cursive, co-founded by the brilliant Talfan Evans. I’d love to see them to set up a lab in Cardiff.
• Hands-on support and mentoring to get entrepreneurs and startups investment ready. And it must come from people who’ve actually done it themselves, not from a workshop deck.
• A joined-up funding landscape making the whole worth more than the sum of its parts. Public money from the likes of the AI Growth Zone, Cardiff Capital Region and the Development Bank of Wales, working in concert with private finance from angels, venture capital and the Wales Pension Funds.
• Utilising our data to enable innovation. In Wales we have the DVLA, Land Registry, Natural Resources Wales, Office of National Statistics, the Patent office to name a few. Imagine if we created sandboxes where the data they hold is anonymised and made available to our academia and businesses. We would be able to create world leading AI innovations that could grow into Unicorns with the support of the previous four points.

The data centres

Which brings me to the bit of physical AI infrastructure everyone wants to talk about: data centres. It’s easy to look at what’s been built in Bristol and get a dose of FOMO. But we need to be honest with ourselves, with investors and with the public about what’s actually possible here, and what we put at risk if we try to jump the queue on our own grid.

The South Wales AI Growth Zone designation, running along the M4 corridor from Newport to Bridgend, is genuinely exciting. A billion pounds of investment, thousands of jobs, the old Ford Bridgend plant repurposed for the digital age. It’s a compelling story. But there’s a problem sitting underneath all of it that isn’t getting anything like enough attention: the South Wales grid cannot currently support AI data centres at the scale being discussed. And once you realise that a new compound semiconductor plant in Newport is often competing with a data centre for the same grid capacity, this stops being a local issue and becomes a question of national strategic planning. Especially when you look at the numbers; an AI data centre employs on average around 20 people at £30k–£40k, whereas a compound semiconductor plant might employ 500 at an average of £68k.

A modern AI data centre campus needs somewhere between 500 megawatts and a gigawatt of uninterruptible power. For context, the entire demand connection for Wales is around 4.5 GW and that’s serving the whole country: industry, homes, existing businesses, everything. The transmission system was built in the 1960s to move power from coal stations to where people lived. It was never designed for dozens of enormous, always-on digital loads clustered along a single motorway corridor. And right now, projects across the UK are being told their grid connection won’t be available until the 2030s.

The grid

So where does new grid capacity for Wales come from? There are two real answers, and both take time.

The first is the Celtic Sea. The potential is enormous; up to 4.5 GW of floating offshore wind connecting into South Wales, enough to power millions of homes. But the honest timeline has that capacity coming online somewhere between 2035 and 2040. And here’s the part that doesn’t get said often enough: one of the main reasons there aren’t more renewable schemes in the South Wales pipeline right now is that the grid constraint is already putting developers off. The grid needs fixing before the renewables can come, and the renewables are needed to justify fixing the grid. It’s a classic chicken-and-egg, and breaking it takes deliberate public intervention.

The second is Wylfa. The decision to site small modular reactors on Anglesey is one of the most significant energy infrastructure calls Wales has seen in a generation, up to 1.5 GW of low-carbon baseload, which is exactly what you’d want sitting underneath a data centre cluster. But Rolls-Royce’s reactor design doesn’t clear its regulatory assessment until late 2026, the final investment decision isn’t expected before 2029, and a realistic grid connection date is the mid-2030s.

Add in the time needed to upgrade the transmission network itself, the new north–south circuit, new substations, the Great Grid Upgrade works and realistically you’re looking at the late 2030s before South Wales has the headroom for large-scale data centres.

In the meantime, we have real choices to make about the capacity we already have. We need to decarbonise heat. Heating accounts for 41% of Wales’s total energy consumption, and shifting that to heat pumps means a lot more electricity demand on distribution networks right across South Wales.

And here’s the point I most want to make. The UK doesn’t need Welsh data centres; it needs UK data centres. The sovereignty argument, the growth argument, the jobs argument none of them actually require the data centre to sit physically in Wales. Scotland has grid capacity to spare, with wind generation regularly exceeding what the network can absorb. Direct data centre investment to the parts of the UK that can support it today and we still capture the national economic benefit. Welsh businesses can still get at the compute, just as they can right now at Isambard in Bristol. What we should be doing is throwing our economic development effort behind the four things I set out at the start of this blog.

Welsh Government’s role

What I’d like to see Welsh Government push for and what NICW is here to inform is a clear sequencing plan. Protect the grid capacity we have for heat decarbonisation and advanced manufacturing. Accelerate the Celtic Sea connection and the Great Grid Upgrade as the foundations everything else depends on. Hold the South Wales AI Growth Zone as a real ambition for the 2030s, switched on as capacity comes online. And in the meantime, make the case loudly in Whitehall that UK AI infrastructure investment should be flowing to grid-ready locations now, with Wales coming into its own once the energy system can genuinely support it.

That isn’t a counsel of caution. It’s a plan for doing this properly and for making sure Wales ends up with both the advanced manufacturing base and the AI infrastructure it deserves, rather than sacrificing one for the premature announcement of the other.