Nebius's £1.7bn UK bet: four AI cloud sites, 65MW, and what builders should take from it

What Nebius just announced

• £1.7bn commitment. Nebius has pledged roughly £1.7bn (around $2.2–2.3bn at current exchange rates) to expand its UK AI infrastructure footprint, announced 8 June 2026 on the opening day of London Tech Week.
• Three new sites, one existing. New Nvidia-powered deployments at Kao Data's Harlow campus, a Park Royal site in west London, and a Greater Manchester site join Nebius's cluster at Ark Data Centres' Longcross Park campus in Surrey, which has been live since November 2025.
• 65MW by 2027. Combined, the four UK sites are expected to reach roughly 65MW of capacity once fully ramped up.
• Private capital, not government money. This is commercial capital from a Nasdaq-listed cloud operator — separate from the UK government's £500m Sovereign AI Fund, which is a different instrument entirely.
• Real customers already on it. Revolut, Prima Mente, Basecamp Research and Sword Health are named as running production workloads on Nebius's UK infrastructure.

Nebius is not a household name the way Amazon, Microsoft or Google are, but in the GPU-cloud world — often shorthanded to "neocloud" — it has become one of the more closely watched operators in Europe. This latest UK commitment is the clearest signal yet that it intends to compete for enterprise AI workloads on British soil, not just sell spare capacity from data centres built for someone else.

Inside the £1.7bn build-out

The headline figure covers four physical sites at different stages of readiness. Only one is actually running GPUs today; the rest are 2026–2027 builds. That distinction matters if you are trying to work out when this capacity is actually usable rather than simply announced.

Capacity and GPU figures per Nebius's and Kao Data's own announcements (8 June 2026) and Nebius's November 2025 UK launch. Combined 65MW figure is Nebius's stated target once all four sites are fully ramped in 2027.

This is not the Sovereign AI Fund's money

It is easy to read "£1.7bn" and "UK AI investment" in the same sentence and assume this is another instalment of the government's £500m Sovereign AI Fund. It is not. The Sovereign AI Fund is a government investment vehicle that takes equity-style stakes in AI startups and scale-ups. Nebius's £1.7bn is private commercial capital, spent by a publicly listed cloud operator building data-centre capacity that it will then sell to customers as a service — the same basic logic as a hyperscaler building a new region, just from a smaller, more specialised operator.

The two threads do intersect at the edges, which is presumably where the confusion starts. Prima Mente, one of the customers Nebius named in its announcement, is itself backed by the Sovereign AI Fund and now trains its biological foundation models on Nebius's infrastructure — so a Sovereign Fund portfolio company is a Nebius customer, without the fund itself having put a pound into Nebius's build-out.

Who's actually building here — and who's using it

Nebius Group is what remains of Yandex N.V. after the Dutch-registered parent sold its Russian assets in July 2024 and renamed itself the following month. Headquartered in Amsterdam and listed on Nasdaq under the ticker NBIS, the company is run by Yandex co-founder Arkady Volozh and has spent the past two years rebuilding a hyperscale-style GPU cloud business largely from scratch, on the back of roughly a thousand former Yandex engineers and infrastructure inherited from the split.

The named UK customers give a sense of what the capacity is actually for. Revolut has rebuilt parts of its AI stack on Nebius, using its Token Factory inference platform for financial-crime detection agents and a customer-support chat orchestrator that Revolut says handles more than a million tickets a month. Pavel Nesterov, Revolut's Executive Director of AI, put it plainly in Nebius's announcement: Nebius "gives us the scale to run training and inference for 75 million customers." Prima Mente is training multi-billion-parameter biological foundation models and needs, in founder Ravi Solanki's words, "sustained, local infrastructure." Basecamp Research and Sword Health round out the named launch customers, in life sciences and digital health respectively.

Nebius vs the neocloud pack

Nebius sits in a category the industry has taken to calling "neoclouds" — GPU-specialist operators built to undercut and out-provision the big three hyperscalers on AI workloads specifically, rather than compete on the full spread of general-purpose cloud services. It is a crowded field, and Nebius is not the largest name in it.

What distinguishes Nebius from the American names in that list is not scale — CoreWeave alone dwarfs it on revenue — but geography and regulatory posture. A European-domiciled, Nasdaq-listed operator building UK and EU capacity gives a British or European enterprise buyer a different procurement conversation than routing the same workload through a US-based provider's international region. Whether that difference is worth paying for is a question every buyer has to answer for their own compliance requirements, not a given.

What this means for Indian builders too

This is a UK infrastructure story first, and it is worth saying that plainly rather than stretching it into something it is not. But it lands the same month India is running its own parallel compute build-out, and the two are worth reading side by side.

On the government side, the IndiaAI Mission has taken shared compute capacity past 34,000 GPUs, offered to eligible startups and researchers at roughly Rs 115–150 an hour — reported at around 40% below open-market rates — with a stated target of 100,000 GPUs by the end of 2026. On the private side, Mumbai-based Neysa closed a Blackstone-led financing of up to $1.2bn in February 2026, aimed at scaling the company from roughly 1,200 live GPUs to more than 20,000 as it builds out what some have called India's answer to CoreWeave — we covered that round in more depth in Neysa's $1.2bn Series B, and previously ran the underlying unit economics against UK pricing in our IndiaAI subsidy vs UK cloud cost comparison. Two different countries, two different funding models — subsidised public compute in India, private commercial capex in the UK — chasing broadly the same outcome: not being entirely dependent on a US hyperscaler's allocation queue for the GPUs your product runs on.

There is also a narrower, more specific reason an Indian-founded team might care about Nebius's UK build even if their engineering team never sets foot in London. If your product serves UK or EU customers — a fintech handling UK account data, a health-tech platform touching NHS-adjacent records — running inference in a UK or EU region rather than routing it back through Mumbai or a US region can be the difference between a straightforward data-residency story and a genuinely difficult one in a due-diligence call. A newly expanded, European-domiciled provider building UK capacity is exactly the kind of infrastructure option a UK-serving Indian startup should be pricing out, regardless of whose government the headline investment figure belongs to.

So — should you actually build on Nebius?

Depends on what you are optimising for.

• Need UK or EU GPU capacity right now: only the Surrey site is live today. The other three are 2026–2027 builds — check real availability for your region and GPU generation before you plan around the announcement rather than the deployment.
• Data residency or compliance is the driver: a European-domiciled, publicly listed operator expanding UK capacity is a reasonable procurement story to bring to a compliance review — but get the specifics confirmed by Nebius's own account team, not from a press release.
• Comparing on cost: neoclouds compete primarily on GPU-hour pricing and contract flexibility, not on the size of their headline investment. Get an actual quote and weigh it against IndiaAI Mission pricing or a hyperscaler UK region before committing.
• You are already shipping on Nebius, CoreWeave, Crusoe or a rival: that is genuinely worth documenting. Infrastructure decisions and the trade-offs behind them are exactly the kind of proof of work that stands out on a Builder profile.

The UK's compute story has moved fast in 2026 — a £500m Sovereign AI Fund, a £400m government chip-buying strategy, AMD's university and startup commitments, and now Nebius's £1.7bn, all inside a matter of months. For builders the practical takeaway is narrower than any of the headlines: work out which of these pots of money, if any, actually changes what GPU capacity you can provision, at what price, and by when — and treat everything else as context.

Nebius's own announcement, with the full list of sites and customers, is at nebius.com/newsroom. Kao Data's confirmation of the Harlow deployment and its capacity figures is at capacityglobal.com.