The UK’s ambitions for a sovereign AI ecosystem are accelerating as Dutch AI infrastructure firm Nebius launches its first UK deployment and private investors pledge billions. Projects from Nscale, CoreWeave, and AI Pathfinder, aligned with government strategies signal a convergence of public policy and private-sector build-out.
In sum – what to know:
Nebius launch – new London-based Nebius facility, powered by Nvidia gubbins, will provide high-performance compute for enterprises and public services, supporting large-scale AI workloads and gen AI development.
Private pledges – investments include Nscale (£2 billion), CoreWeave (£2.5 billion total), and AI Pathfinder (£18.4 billion), reflecting strong commercial alignment with government AI infrastructure policies.
Government sovereignty – government initiatives like the AI Opportunities Action Plan, AI Growth Zones, and Compute Roadmap guide seeks to address the country’s AI infrastructure gap.
The move by Dutch AI infrastructure firm Nebius to deploy its first UK-based infrastructure, built on Nvidia Blackwell Ultra GPUs, is a further sign that the UK’s ambitions for a domestic AI ecosystem are gathering a head of steam. It follows in line with the UK government’s AI Opportunities Action Plan, published in January, and suggests public-sector policy and private-sector build-out are starting to converge quite well in pursuit of an AI goldmine.
The plan made “sufficient, secure, and sustainable AI infrastructure” a central plank of the UK’s economic strategy, and set goals to expand its sovereign AI Research Resource (AIRR), a network of publicly-funded high-performance computing (HPC) clusters, presented as a national compute infrastructure for AI, by at least 20-fold by 2030, and to establish new “AI Growth Zones” as regions with fast-tracked AI infrastructure planning and power access
In tandem, the AI industry has responded. As well as the new Nebius project, UK-based Nscale, a self proclaimed AI infrastructure “hyperscaler”, announced shortly after the government package that it would invest £2 billion ($2.5 billion) to build UK data-centre capacity, via acquisition of a site in Loughton in Essex with 50 MW (scalable to 90 MW) of power capacity, tailored for AI compute.
Again, Nvidia is the go-to GPU supplier, and the whole project puts emphasis on sovereign AI. Nvidia, again; plus the announcement made reference to closed-loop cooling and renewable energy in the name of sustainability. As well, CoreWeave has just (September) confirmed an additional £1.5 billion in the UK to “power the next wave of AI innovation by building facilities”, bringing its total investment in the country to £2.5 billion.
The CoreWeave project is explicitly tied to supporting the UK Government’s Compute Roadmap, a distinct foundational policy document issued by the Department for Science, Innovation and Technology (DSIT) and UK Research and Innovation (UKRI) in July to complement its AI action plan, with a focus on national compute infrastructure, public and private, for science, innovation, and AI.
Besides, UK firm AI Pathfinder revealed in September that it will invest £18.4 billion in two mega-sized UK data centres: one in Northamptonshire (£3.4 billion) and one in Ayrshire, Scotland (£15 billion). The Northamptonshire site will be operational by December, apparently. The scale of investment suggests an ‘AI-factory’ style deployment, and Data Center Dynamics references Dell and NVIDIA hardware for large-model training and inference.
The UK government has said at least £25 billion of private-sector investment in new UK data centres has been announced since July. It is putting-up, as well: its AIRR facilities will receive £1 billion of state money. Its own policy-linked private investment vehicles, as above, are geared to attract investment and also create enabling frameworks for infrastructure deployment. A new ‘growth zone’ in the North East could attract up to £30 billion in private investment and create more than 5,000 jobs, it reckons.
Of course, headlines only tend to get attached to projects when they are announced; the proof will be in how many become operational, on schedule (through 2026-27), as well as whether the new infrastructure is used for sovereign high-value AI builds, such as foundation models, or mostly just for smaller scale workloads. Many announcements are “commitments”, as well; the key will be how much capacity gets used, particularly by enterprises, representing the rest of the economy.
But the new Newbius launch is worth considering. Its deployment, already live, combines “supercomputer power and hyperscaler simplicity”, it says, via local specialised AI compute. The hardware includes NVIDIA Blackwell Ultra GPUs and NVIDIA Quantum-X800 InfiniBand networking. “We’re empowering organisations across the UK to train, deploy, and scale AI models and applications faster, more securely, and more sustainably than ever,” it states.
The facility, located in the London region, is built with energy-efficient cooling, low-latency networking, resilient on-site power, and is integrated with the NVIDIA AI Enterprise software stack. UK customers across startups, enterprise, academia, and public services (including the NHS) will be able to access it. It is significant insofar as it adds new domestic capacity for large-scale AI workloads, and supports generative AI and foundation-model development, which require extreme GPU/InfiniBand scale.
As above, it also aligns with policy goals around data sovereignty, security, and UK-based infrastructure. It is part of a global roll-out by Nebius but brings the capability to the UK market now.
The UK remains among the leading AI markets globally in research talent, but infrastructure has been flagged as a gap. Most notably, perhaps, Nvidia chief Jensen Huang remarked on a panel with prime minister Keir Starmer in June that the UK is in a “Goldilocks circumstance”. He said: “The ecosystem is really perfect for takeoff — it’s just missing one thing.” The one thing is infrastructure, of course.
Some of Nvidia’s biggest customers, including Nebius, are looking to put that right.
