Fuse Energy, a London electricity supplier founded in 2022 by former Revolut executives Alan Chang and Charles Orr, has opened reservations for Fuse Compute, which gives companies access to NVIDIA GPUs in UK data centres that run on Fuse’s power.
The company supplies 470,000 UK homes, and Fuse Compute is its first product in commercial AI infrastructure.
Customers rent bare-metal, single-tenant nodes, meaning each company gets whole machines instead of sharing hardware with other tenants, and the service runs under UK jurisdiction with a UK data path. Pricing, terms and delivery are agreed per contract.
The first 128 NVIDIA B300 GPUs are due in December 2026, and Fuse expects to have 1,024 GPUs by the second quarter of 2027, 22MW of data centre capacity in the fourth quarter of 2027 and 116MW from 2028.
To support this plan, the company points to a data centre connection pipeline of 142MW across 15 projects and a power generation pipeline of more than 1GW.
Chang explained the reasoning in a statement accompanying the announcement: “Power is the biggest hurdle to building AI infrastructure, and it’s the part we already know how to do.”
The UK needs 6GW of AI capacity while data centres queue for power
The government’s compute roadmap, published in July 2025, forecasts that the UK will need at least 6GW of AI-capable data centre capacity by 2030. That is three times the data centre capacity available when the paper was published.
In the same document, the government says it will support the development of behind-the-meter, low-carbon energy solutions, which supply a site directly instead of drawing power through the grid.
The Treasury published the Data Centre Covenant alongside the Budget on October 28, and it sets out further measures, including reserving grid capacity for strategic data centres, transferring pre-2030 connection contracts to them, and allowing exceptionally strategic projects to use tightly controlled on-site generation until they can secure a grid connection.
The government needs these measures because the queue for connections is long. In February, The Register reported that the National Energy System Operator had identified about 140 data centres in the demand queue, with a combined demand of 50GW. At that point, the highest electricity demand Britain had recorded in 2026 was 45GW.
Fuse says new data centres can wait up to a decade for a connection.
Microsoft’s exit from Loughton shows what grid delays cost AI companies
The effect of that queue became visible in October, when Microsoft pulled out as anchor customer of the data centre Nscale is building in Loughton, Essex, a project Nscale had pitched as the UK’s largest sovereign AI data centre.
Microsoft had agreed to rent 23,040 NVIDIA GB300 GPUs at the site from the first quarter of 2027, and it said only that its decision concerned a single site. Grid officials, however, had told Nscale they did not expect to supply power to Loughton before the 2030s, while the site had originally been due to open in the fourth quarter of 2026.
The delay came from the wider network rather than from the site itself. According to The Guardian, UK Power Networks had warned Nscale the capacity it needed might not arrive until the 2030s, and a spokesperson for the network operator said many data centres “are awaiting works on the national transmission network which we do not own.”
Nscale described the problem as an “unfortunate London-wide grid issue” and said it did not affect its other contracts. For an AI company deciding where to rent compute, the case shows that GPUs and a large customer can both be secured while the opening date still depends on grid works that neither the developer nor its customers control.
Nscale’s response points to the same factor Fuse is building its offer around. The company said it is exploring alternative power sources for Loughton, and Capacity has reported talks with Bloom Energy about on-site fuel cells, an option that adds cost and planning complexity.
The government’s covenant allows on-site generation only for exceptionally strategic projects and only until a grid connection is secured, so most developers still have to wait their turn. Fuse is presenting its experience in generation and its connection pipeline as a way to avoid that wait, and that is the basis on which it is asking AI companies to choose it.
Fuse is selling speed and UK jurisdiction rather than lower prices
Chang has said that Fuse does not expect cheaper energy to make its GPUs cheaper. He told City AM that “energy costs are only five per cent of GPU rental revenues, so energy costs are not a significant consideration when pricing GPU rentals,” and he said Fuse would set competitive prices without committing to charge less than established providers.
“The key advantage Fuse possesses is energy availability, which means much faster speed to stand up data centres, and this will allow our compute customers to scale up much faster,” he said on Fuse’s competitive advantage.
Jurisdiction is the second difference between Fuse and the large U.S. cloud providers, since those providers fall under the CLOUD Act, the 2018 law that allows the U.S. government to demand data from American tech companies even when the servers sit outside the United States.
Microsoft France’s director of public and legal affairs was asked about this under oath at a French Senate inquiry in June 2025, and he said he could not guarantee that data on French citizens would stay out of U.S. government hands without the consent of French authorities.
A UK company renting single-tenant machines in UK data centres gives AI companies that handle sensitive data a clearer answer about where their workloads run. Chang also pointed to latency, saying AI companies based abroad could run their British services from Fuse’s facilities to reach UK users faster, which makes the offer relevant to European companies with customers in the UK.
How much Fuse adds to the 6GW target
Fuse’s 142MW pipeline amounts to about 2.4% of the 6GW the government says the UK needs by 2030. That share is a 150sec calculation, which divides Fuse’s stated data centre connection pipeline by the government’s 2030 forecast.
The 128 GPUs due in December are also a small number next to the 23,040 that were planned for Loughton. Fuse’s plans will not close the national gap by themselves, so their value for AI companies depends on whether the capacity arrives on the dates Fuse has set.
Fuse is also preparing to use NVIDIA hardware for its own operations. In October, the company announced that the AI agent handling customer support in its app will switch from closed-source models to open-weight models hosted on NVIDIA systems that Fuse controls.
The agent has an 80% self-containment rate, which means it settles that share of queries without a human agent stepping in, and a reported 99% customer satisfaction score that Fuse expects to keep after the change.
Running this workload in-house gives Fuse experience in operating GPU infrastructure for its own business before it delivers capacity to paying customers.
The first 128 GPUs are due in December
Whether controlling power shortens the time it takes to bring GPUs online will become clear in December, when the first B300s are due to go live.
The 142MW Fuse points to is a pipeline of connections across 15 projects, not capacity already built, and a generation pipeline of more than 1GW does not connect a data centre by itself.
If Fuse meets its December and mid-2027 dates while projects such as Loughton are still waiting for power, AI companies in the UK will have evidence that a supplier which already works on generation and grid connections can deliver compute sooner than developers that depend on the queue.
Featured image: via Fuse
