Will the UK be an AI maker, or an AI taker?
Right now, data centres are fashionable in the UK. September’s State Visit of President Trump coincided with a flurry of announcements: billions invested to develop AI in Britain, and with it data centre infrastructure.
Industry giants Microsoft, Google, NVIDIA, Open AI and CoreWeave are investing a total of £31 billion to boost the UK’s AI infrastructure with the help of home-grown partners. British company Nscale is working with OpenAI to deliver the Stargate UK project, and with Microsoft to deliver an AI supercomputer in Loughton. Google is opening a new data centre in Waltham Cross, Hertfordshire. CoreWeave is working with British firm DataVita in Scotland to build one of Europe’s “largest, most efficient” AI data centres. And global investor BlackRock is also pumping £500 million into enterprise data centres across the country, including an initial investment of over £100 million in a data centre expansion west of London.
All of which is good for UK plc. “We're delighted that the big tech firms in the US made these announcements,” enthuses Spencer Lamb, chief commercial officer at data centre developer Kao Data. He expects the various deals to help boost Kao’s Harlow campus, which will eventually comprise four 10MW data centres. Other Kao Data data centres in Manchester and Park Royal, London are also set to benefit, Lamb adds.
But while the investments are “great news”, he suggests they are “not quite as significant as the UK government set out”. That’s because the huge volume of new processing power (NVIDIA is deploying a total of 120,000 advanced GPUs across the UK, said to be the largest rollout of its type in Europe) will be used to train the AI and host the cloud resources of US companies, thus “monetising the models that are being created in America”, says Lamb.
“The ambition the government set out at the beginning of the year was to create a situation where the UK becomes an AI maker, not taker.
“We're still taking, we're not actually making.”
Norway power prices spur data gigafactory
In the summer, Nscale made another announcement: the launch of another Stargate, in Narvik, northern Norway: a giant advanced AI infrastructure project developed in partnership with Open AI and Aker, which will in itself deploy 100,000 NVIDIA GPUs by the end of 2026, or 230MW of capacity (with the potential to expand to 290MW, all powered by hydropower). “Power prices well below the European average” and a “cool climate” were part of the rationale for the project, says Nscale.
As Lamb points out, a project of the scale of the Norwegian scheme dwarfs any single development in Britain, even with the positivity of that US-UK pact on AI. “It's clear to me that we haven't quite achieved that ambition of having large AI clusters where models are created by our bright young things: the people who are able to create the next ChatGPT.”
Lamb says that hyperscalers experience “two key barriers to them deploying much greater data centre capacity”. One of these is the cost of energy; the other is copyright law, which Lamb believes should be revised along EU lines.
“The cost of energy is so significant in the UK,” he says. Lamb adds that high-energy users in Germany enjoy discounts to non-commodity costs which help make data centres more competitive. “Our electricity cost for data centres is probably double what it is in Europe, and three or four times what it is in America and in the Nordics.
“When you talk to the likes of Open AI, they say copyright law and high energy costs are big barriers.”
Levelling up the regions with data centres?
Currently the data centre market in the UK is focused on west London. Lamb says lower energy prices and tapping into abundant renewable power have the potential to develop data centres in the regions. “If power costs and copyright law were resolved, we would see an inflection point in the data centre industry in the UK where US tech firms would be happy to put their AI training data centres into Scotland and Northeast England in a really meaningful way: you would create a North-South data centre configuration.” (The government claims a new “AI Growth Zone” in the Northeast has “the potential to see billions of pounds worth of investment and jobs funnelled into the region”; for its part, Kao Data is hoping to develop a £75 billion data centre cluster on the site of an old steelworks between Glasgow and Edinburgh with the potential for 500MW of capacity).
Another barrier to data centre development is the queue for new connections to the grid. This is a particular problem in “constrained geographies” such as the southeast, says Lamb. “There’s not much the networks can do to accelerate that process.
They might be able to shave a bit of time off, but I don't get the sense there's an awful lot that more they can do.”
One answer would be to develop power infrastructure in the regions in partnership with independent private sector businesses to accelerate substation build in tandem with transmission owners. “We could potentially encourage a… structure on the transmission network whereby it becomes far more competitive by getting private companies to upgrade the infrastructure, rather than rather just relying on national groups,” says Lamb. Such a move would replicate the development of independent nodes on the distribution network, he adds.
Either way, the next few years are likely to prove crucial to data centre development in Britain. “Between 2025 and 2030 is where the global AI race is going to be won. What we need to do is deploy data centre capacity between now and then so we've got large-scale AI models being trained, monetised and exported.”
Get that right, and Britain can be an AI maker – as well as the beneficiary of inward investment from the US.
“We’re positioned third in the global AI arms race currently. If we don’t do this, we will drop down that list,” Lamb concludes.