Has the UK’s AI infrastructure buildout been a success?

When the U.K. announced its AI Opportunities Action Plan — a grand blueprint to deploy the tech across society — in January, Prime Minister Keir Starmer declared the strategy would make the country an “AI superpower.” 

One of the key pillars of this plan was a rapid buildout of data centres capable of providing the huge compute requirements for the rollout of AI. This would be driven by “AI growth zones” — designated areas with relaxed planning permission and improved access to power. 

Nearly one year on, and Nvidia, Microsoft, and Google have all committed billions of dollars to AI infrastructure in the country. Four AI growth zones have been unveiled, and homegrown startups like Nscale have emerged as key players in the space. 

But critics point to heavily restricted access to energy via the national grid and slow-moving buildouts as signs the country is at risk of lagging further behind global rivals in the AI race. 

“Ambition and delivery are not yet aligned,” Ben Pritchard, CEO of data center power supplier AVK, told CNBC. 

“Growth has been held back largely by constraints around power availability. Grid bottlenecks, in particular, have slowed the pace of development and mean the U.K. is not yet deploying infrastructure quickly enough to keep pace with global competitors.”

It is still early days in the U.K.’s AI infrastructure buildout as AI growth zones are currently in their initial phases of development.

A site in Oxfordshire, the first to be announced in February, has yet to begin building work and is still considering delivery partner proposals. Ground preparation work has begun at one in the North East of England, announced in September, with formal building beginning early 2026.

Two more sites, in North and South Wales were unveiled in November. The former is searching for an investment partner, which the Department of Science, Technology and Innovation (DSIT) told CNBC it expects to be confirmed in the coming months. The latter is made up of a cluster of sites, some of which are already operational with additional construction work to be done on others, DSIT said.

The U.K. government said in July it was targeting a core group of AI growth zones serving at least 500 megawatts of demand by 2030, with at least one scaling to more than one gigawatt by that time.

But the most serious challenge to realising those ambitions is the U.K.’s limited grid capacity, said Pritchard.

“Developers expect grid connection delays of eight to ten years, and the volume of outstanding connection requests, especially around London, is unprecedented,” he told CNBC. 

AI workloads are also “dramatically increasing energy demand” as businesses and consumers begin to use the tech, putting additional pressure on a stretched energy system, Pritchard added. “They are no longer isolated risks; they are actively slowing down or blocking developments across the country.”

The open call for applications for the AI growth zone initiative created a situation where landowners with pylons or powercables running across their land applied for the designation, said Kao Data’s Spencer Lamb.

“This resulted in the national grid being inundated with power grid applications from speculative sources,” with no realistic chance of success, he told CNBC.

The National Energy System Operator (Neso) — The U.K.’s public body responsible for managing the national grid — has made moves to fix the situation. 

Earlier this month it announced plans to prioritise hundreds of projects for faster access to the grid. Neso declined to comment on whether AI infrastructure projects were among those prioritised when asked by CNBC, but did say a significant portion were data centers.

There have also been big money commitments from tech giants, many of which were paraded by the U.K. government in September.

Microsoft, Nvidia, Google, OpenAI, CoreWeave and others announced billions of dollars of AI investment during U.S. President Donald Trump’s state visit, which involved plans to deploy the latest chips in the country and open new data centers. 

Homegrown startup Nscale, which provides access to AI compute and is building data centers, also announced deals to deploy tens of thousands of Nvidia chips at an AI factory just outside London by early 2027.

“Investment from major private players has laid important groundwork,” Puneet Gupta, general manager for the U.K. and Ireland at data infrastructure company NetApp, told CNBC. “Momentum is also building around national research supercomputers and plans for new compute capacity, with commitments to build AI ‘gigafactories’ in the UK.”

But the “real test” will be how quickly those plans translate into usable compute for U.K. organisations, said Gupta.

The long-term success of the country’s AI infrastructure buildout will require it to invest in the “full stack,” including data pipelines, storage, energy sourcing, security, talent and skills, Stuart Abbott, U.K. and Ireland’s managing director at AI infrastructure company VAST Data, told CNBC. 

That means “developing an operational fabric that lets real institutions deploy AI safely at scale,” he added. “If the UK wants this to be durable rather than a one-year sugar rush, it has to treat AI infrastructure like economic infrastructure.”

The challenges are significant. The value of data center deals in Europe pales in comparison to sums funneled into projects in the U.S. The U.K. also currently has the costliest energy in Europe, which is around 75% higher than before Russia’s invasion of Ukraine, and legacy grid infrastructure which can take many years to connect to new sites.

One potential solution for projects that are unable to secure access to the national grid are microgrids, AVK’s Pritchard said. Microgrids are self-contained power networks from sources like engines, renewables and batteries. 

AVK is currently designing two microgrids for partners building cloud compute, though not for AI, in the U.K. They can take around three years to build and cost around 10% more than energy from the grid at the moment, according to Pritchard. 

Co-locating compute where power already exists, rather than “forcing everything to be greenfield” — the term for undeveloped sites — is also a way to get AI infrastructure up and running faster, VAST Data’s Abbot said.

The pace of implementation will be critical, Kao Data’s Lamb told CNBC. “Unless fundamental issues around energy availability and pricing, AI copyright and funding for AI developments are solved quickly, the U.K. will miss out on one of the most remarkable economic opportunities of our time and ultimately risks becoming an international AI backwater.”