White paper: We can’t wait on fusion for the power AI needs

Written by
Simon Ordish
|
November 16, 2026
|
6
Microsoft’s reported exit from the Loughton data centre shows what happens when power isn’t ready. This paper looks at four practical ways to power Britain’s AI build-out in the 2030s and why fusion is a question for later.

Let’s be realistic: large-scale fusion is exciting, and may be the answer to AI’s ever-growing appetite for power, but it’s a way off yet. Scaling AI computing and data centres through the next decade needs a more pragmatic solution. Put the infrastructure where the power is now. Grow the transmission network to open up new sites as quickly as possible. And bridge any gaps with on-site generation. It’s not glamorous, but it will work. 

We’ve just seen what happens with a build-first, connect-later mindset. In October 2026, Microsoft withdrew as the anchor tenant from Nscale’s planned AI data centre in Loughton, Essex. Neither party gave a formal reason for the split, but it came shortly after grid operators warned that the power supply the site needs would likely not be in place until the 2030s. Even if that’s an unfortunate coincidence, it’s an illustrative one, as we discuss in our white paper Ready to build, but waiting to connect.

Make no mistake: fusion is an important part of the long-term story, and the UK is well positioned to take advantage. But we’re short of power today. For now, scaling the infrastructure we need, and getting power to it, both depend on the same thing: the ability to deliver large, complex infrastructure on time and on budget.

What’s at stake for the UK?

To data centre investors and developers, the UK doesn’t look particularly attractive at the moment. Two flagship AI projects announced around the 2025 US state visit have stalled within six months of each other. In April 2026, OpenAI paused its Stargate UK project, commenting that high electricity costs and uncertainty over copyright rules made it “not feasible for the time being”. And as mentioned above, in October, Loughton lost its anchor tenant after a significant grid-connection delay was announced.

Energy prices don’t help the overall picture. UK large industrial users paid 25.33 p/kWh in the first half of 2025, roughly 123% above the EU-14 median. Global rankings frequently place the UK among the top five most expensive developed nations.

It would be unfair to say these challenges are unique to the UK. But the grid is the root cause the country has control over. And if that bottleneck isn’t solved, hyperscalers will increasingly look elsewhere. AI compute is far more mobile than traditional manufacturing, engineering or logistics businesses. If data centres can be connected faster elsewhere, to cheaper power sources, the sector will vote with its feet.

Four ways forward without fusion

The UK does not lack electricity in aggregate. It lacks electricity in the right places, on the right terms, at the right time. There are four ways to address that, and none requires a major scientific breakthrough.

  • Build where the power already is. In 2025, Great Britain switched off a record 10TWh of wind generation, with more than 8.8TWh of it in northern Scotland, at a cost of about £1.4bn. Data centres are among the few loads large and steady enough to soak up that power locally – and four of the five AI Growth Zones listed by government are in the North East, Wales and Scotland.

  • Connect on flexible terms. A firm connection that waits a decade is worth less than a flexible one available now. One report puts the share of data centre projects that would accept non-firm or phased connections for an earlier date at 81%, and DESNZ is considering a mandatory minimum level of demand flexibility for very large users. That turns the data centre from a burden on the grid into a resource the system operator can call on under stress.

  • Bridge with on-site generation. Ofgem reports about 117 active cases and enquiries from data centres seeking gas network connections. Nscale says it is exploring alternative power at Loughton and has reportedly held talks with a fuel cell supplier. Bridging power is not the end state, but it separates the opening date from the connection date, provided it is designed to step down cleanly when the grid arrives.

  • Deliver the transmission build on time. RIIO-T3 and the Accelerated Strategic Transmission Investment programme are funding the largest expansion of the GB network in decades. Every circuit and substation delivered releases capacity. Every one that runs late pushes a connection date back somewhere else. This is the UK's biggest lever, and delivery challenge.

When can we expect a fusion solution to AI energy needs?

The hyperscalers are clearly taking fusion seriously. Microsoft signed the world’s first fusion power purchase agreement with Helion in 2023, for up to 50MW from a plant in Washington State, with delivery targeted for 2028. Google has signed for 200MW from Commonwealth Fusion Systems, with first power projected for the early 2030s.

But it’s important to understand the purpose behind these deals. No fusion device has yet produced more electricity than it consumes. One independent academic said it would be surprising if Helion delivered power routinely in 2028, and Microsoft itself says fusion is “by no means the only solution” it supports. These companies are investing in fusion as a long-term option, not a credible short-term supply.

The UK’s own programme is upfront about the expected timescales. STEP Fusion, the prototype plant at West Burton in Nottinghamshire, is working towards construction in the early 2030s and first operations in the early 2040s, backed by £1.3bn within a wider £2.5bn for fusion.

So fusion isn't the answer for Loughton, for Stargate, or for any programme commissioned this side of the 2040s, at best. It’s a credible future technology that might one day meet the UK’s power needs (whatever they may be in a decade or two). It also relies on more than just physics, as we discussed in our white paper: It’s not just atoms that need bringing together in UK Fusion.

The same problem, today and in twenty years

Fusion might eventually solve the cost problem that keeps the UK near the top of the list for energy prices. But it won't solve the connectivity problem. Fusion reactors will be large, centralised facilities, and unless we build all our data centres next door, we’ll still need sufficient grid capacity and rapid connectivity to solve the problems discussed above. 

Getting the grid right now is therefore an investment in the future as well as a short-term fix. In the meantime, if you’re navigating Ofgem’s new regulations on connectivity, planning a build that depends on a grid connection being ready at handover, or supplying into STEP, talk to Majenta about testing the sequence and the information plan before they are tested on site.   

Key contact

Priya Flora, Partnership Manager, Majenta Solutions

Sources:  

‍