Fuse Energy Takes on Britain’s AI Power Problem With Its Own GPU Cloud

Britain has plenty of ambition for artificial intelligence, but it has a stubborn problem that sits underneath all of it: getting enough electricity to the right buildings at the right time. Data centres can wait years for a grid connection, and that wait slows down everyone from startups training models to hospitals testing new tools. London energy company Fuse Energy thinks it can shorten that wait by doing both jobs at once, supplying power and running the computers that use it.

The company has launched Fuse Compute, a new service that will offer Nvidia graphics processors hosted in UK data centres and powered by Fuse itself. According to reports published on 21 October 2026, the first 128 Nvidia B300 GPUs are due to come online in December 2026, rising to 1,024 in the second quarter of 2027. Later phases are planned to reach 22 megawatts of capacity in the final quarter of 2027 and 116 megawatts from 2028. For a country that has talked a lot about AI sovereignty, this is a practical test of a simple idea: if power is the bottleneck, own the power.

Why power became the AI bottleneck

For years, the public conversation about AI focused on chips, models and talent. Those still matter, but they are not the constraint that keeps executives awake in Britain. The UK Government's Compute Roadmap, cited in reporting on the launch, estimates the country will need at least 6 gigawatts of AI-capable data centre capacity by 2030. At the same time, new data centres can face waits of up to a decade for a grid connection.

That gap explains why a power company entering the compute market is not as strange as it first sounds. Building a data centre is only half the job. Energising it, at the scale AI workloads demand, is often the harder half. Fuse says its most recent solar site was built at around 30 percent below the industry average cost per megawatt, and it holds a power generation pipeline of more than 1 gigawatt, alongside a data centre connection pipeline of 142 megawatts across 15 projects. Those numbers, reported by the company, are the core of its pitch: it already does the hard part, so adding computers is a natural next step.

The UK Government has set a target to expand public compute capacity twentyfold by 2030, yet many British AI companies still rely on capacity hosted overseas.

What Fuse Compute will actually offer

Fuse Compute is described as a bare-metal, single-tenant service. In plain terms, that means customers get dedicated GPU nodes rather than sharing hardware with strangers through a virtual layer. For companies running sensitive workloads, or tuning performance closely, that dedicated setup can be easier to reason about than a shared cloud instance.

The service will run Nvidia B300 GPUs, part of Nvidia's Blackwell Ultra family, in UK data centres, with a UK data path and operations under UK jurisdiction. Fuse says it is already taking reservations. The first phase is small by hyperscale standards: 128 GPUs in December 2026, with 700 kilowatts live that month according to the company's own site, rising to 3.25 megawatts in the second quarter of 2027.

Fuse is not a newcomer to energy supply. It supplies around 470,000 UK homes and holds both supplier and meter operator licences. It owns and operates low carbon generation assets, with 22 megawatts active and a pipeline above 1 gigawatt across the UK, the United States and Europe. That background is what separates this launch from a typical cloud announcement. Fuse is not renting power and hoping the grid delivers. It is betting that generating power, and holding connections, gives it a head start.

Why this launch fits a wider UK pattern

Fuse is not the only British company trying to solve the power puzzle. What makes it different is the direction of travel. Most compute companies start with servers and then go looking for power. Fuse starts with power and adds servers.

By hosting in the UK, Fuse says British AI companies will no longer need to rely on overseas capacity. For organisations handling sensitive information, domestic infrastructure can be a genuine consideration when choosing a provider.

Limits remain. Reporting notes that details about the specific data centres, and who operates them, have not been fully provided, and prices will need to be negotiated individually. And 128 GPUs is a long way from the gigawatt scale the national roadmap describes. The question is whether Fuse can repeat this trick, site after site, without hitting the same grid delays everyone else faces.

What to watch next

The first test arrives in December 2026, when those initial GPUs are due online. If they arrive on schedule, Fuse will have shown it can move from announcement to operating hardware in weeks, not years. The second test comes in the second quarter of 2027, when the count is planned to jump to 1,024 GPUs. That eightfold step up will reveal whether the connection pipeline is as deep as claimed.

For now, Fuse Compute is best read as a sign of where the UK AI race is heading. The fight is no longer only about who has the cleverest model. It is about who can get power, land and cooling lined up in the same place at the same time.

Frequently Asked Questions

What is Fuse Compute?

Fuse Compute is a new AI infrastructure service from UK energy company Fuse Energy. It will offer dedicated Nvidia GPU nodes hosted in UK data centres and powered by Fuse, starting with 128 Nvidia B300 GPUs due online in December 2026.

How many GPUs will Fuse Compute offer, and when?

According to reports on the launch, the plan starts with 128 Nvidia B300 GPUs in December 2026, rising to 1,024 in the second quarter of 2027, with further capacity of 22 megawatts planned for the final quarter of 2027 and 116 megawatts from 2028.

Why is a power company launching a GPU cloud?

Fuse argues that power is now the main constraint on AI data centre growth in the UK, with grid connection waits of up to a decade reported for new sites. Because it already generates power and holds a pipeline of data centre connections, it believes it can bring AI compute online faster than rivals that must secure power first.