Why does it make sense for Nebius to launch in the UK ?
The first push to become the cloud of reference for sovereign AI
When Nebius announced they’re bringing NVIDIA B300s to the UK, my first reaction was excitement. And the more I looked at the UK specifically the more interesting it became. Because hidden beneath the surface of what looks like a straightforward geographic expansion for the king of neoclouds in the EU is actually a remarkably well-timed bet on a market that’s over-served while also being desperately under-supplied.
The GPU shortage is still a thing
AI investment in the UK hit a record £2.9 billion in 2024, with over 5,800 AI companies now operating in the country, representing an increase of 85% since 2022. That’s real money flowing into real companies that are trying to build real products.
And the problem is that most of them can’t get the compute they need.
Sure, the H100 shortage “ended” in 2024. Lead times dropped from 11 months down to 2-3 months by mid-2024. Everyone moved on to worrying about the next bottleneck : power, cooling, whatever.
But actually the shortage didn’t end because supply caught up with demand. It “ended” because hyperscalers like AWS made it easier to rent GPUs for shorter periods, and some companies started reselling their excess inventory. So, the underlying demand is still way bigger than the supply, especially for companies that aren’t named OpenAI or Anthropic.
Of course, we could have guessed this from the earning calls of hyperscalers and neoclouds throughout 2025.
However in the UK, this supply-demand imbalance is even more pronounced. The hyperscalers are there, sure, but they’re globally constrained. Their UK capacity is a fraction of what they have in US regions. And when push comes to shove, when a major foundation model lab in Silicon Valley needs another cluster, where do you think those GPUs are going?
Not to the fintech startup in London trying to build better fraud detection. Not to the biotech company in Cambridge training drug discovery models. Not to the hedge fund that needs on-demand access for backtesting strategies.
This is where Nebius walks into a real opportunity. Because specialist GPU cloud providers aren’t just competing on having the chips, they’re competing on actually making them available to customers who can’t command the attention of hyperscaler account teams.
This has been the true Nebius moat ; and something that they have been insisting on the past few quarters : they care as much for the 100B+ client and for the small startup.
Why UK, Why Now
The UK attracted more than £14 billion in AI-related investment in just the six months leading up to January 2025, with an average of £200 million per day in private sector investment flowing into the AI sector. More than big numbers, that’s momentum. That’s companies raising money specifically to build AI products and needing somewhere to train and deploy them.
But let’s zoom out for a second and think about Europe more broadly. If you’re Nebius and you want to push sovereign AI, where do you want to push investements?
France? Decent AI scene. Macron has been pushing the AI narrative hard, and they already have a small datacenter in operation near paris. But again: language, and you’re dealing with the full force of EU regulation, particularly the AI Act that’s now in effect. The Netherlands? Actually not a bad choice. Amsterdam is business-friendly, English-fluent, has good infrastructure. But it’s a smaller market.
The UK, though? The UK is different. And Brexit—controversial as it is—actually matters here in ways that most people aren’t thinking about.
The UK has explicitly stated it has “freedom now in relation to regulation to do it in a way that we think is best for the U.K.” after Brexit, choosing its own path rather than following either the EU or US approach. While the EU AI Act’s extraterritorial scope means UK companies operating in EU markets must still comply, for infrastructure providers serving primarily UK customers, the regulatory landscape is meaningfully lighter.
Think about what this means for a GPU cloud provider. The EU AI Act has specific requirements for high-risk AI systems. It mandates certain levels of transparency, human oversight, safety obligations. If you’re running infrastructure in an EU member state, you’re subject to all of that. You need to understand how your customers are using your compute. You need documentation. You need compliance frameworks.
The UK are taking what they call a “pro-innovation” approach. Sector-specific regulation, not horizontal AI legislation, which is much less prescriptive.
For Nebius, this is huge. They can move faster and serve customers who are building cutting-edge AI applications without getting tangled in compliance requirements that are still being figured out in Brussels. And critically, they can position themselves as the compute provider for UK companies that want to operate in a more flexible regulatory environment while still having access to European markets.
The Competitive Moat Nobody’s Built Yet
Let’s talk about CoreWeave for a minute, because they’re the obvious comparison. CoreWeave opened two UK data centers in late 2024, part of an initial £1 billion investment. They’re in Crawley and London Docklands. They’re running some of Europe’s largest NVIDIA H200 deployments. They’ve got first-mover advantage.
But here’s the thing: first-mover advantage in infrastructure isn’t as durable as people think. What matters is execution, availability, and customer service. And right now, in late 2024 and early 2025, the UK market is nowhere near saturated.
CoreWeave alone cannot serve this market. British AI companies now contribute £11.8 billion to the UK economy (double the amount from 2023) with AI employment topping 86,000 across the country. The compute demand implicit in those numbers is staggering.
And here’s what I keep coming back to: Nebius isn’t just another GPU cloud provider throwing chips at the problem. They’re a company that was born out of Yandex’s infrastructure division. They built and operated one of the largest cloud infrastructures in Russia. Say what you want about Yandex, but they knew how to run infrastructure at scale in challenging environments.
That expertise translates. Operating GPU clusters is about cooling, power management, network optimization, job scheduling, customer support. It’s about building a platform that AI engineers actually want to use, not just tolerate because they have no other option.
The UK market right now reminds me of the early cloud wars in 2010-2012. AWS had launched. Microsoft and Google were scrambling to catch up. But there was enough demand that everyone could grow. The market was expanding faster than any one provider could capture it.
We’re in that phase now for GPU compute in the UK. And Nebius is positioning itself to be one of the two or three providers that actually matter in this market.
The Financial Services
Here’s something I haven’t seen anyone else discuss: London is still one of the world’s premier financial centers. And financial services firms are some of the most aggressive adopters of AI that nobody talks about.
Banks, hedge funds, insurance companies… They’re all building proprietary trading algorithms / risk models / fraud detection systems / customer analytics platforms. And they have specific requirements that make hyperscalers a poor fit.
First, latency. If you’re running real-time trading algorithms, you need your compute physically close to where the trading happens. A few milliseconds of latency can mean millions of dollars. You can’t be routing traffic to a US East data center.
Second, data sovereignty. UK financial regulators care deeply about where data is processed and stored. Having UK-based compute infrastructure eliminates a entire category of regulatory questions. You’re not shipping sensitive financial data across borders. You’re keeping everything in-country.
Third, and this is subtle: financial firms want providers who understand their security requirements. They want to be able to audit your infrastructure. They want private clusters, not shared multi-tenant systems. They want guarantees about who else is running on the same physical hardware.
Nebius, coming from the Yandex lineage, has experience serving enterprise customers with stringent requirements. They’re not optimizing for consumer internet scale. They’re optimizing for customers who will pay premium prices for guaranteed availability and security.
And in the UK financial services sector, there’s deep pockets for that kind of service. Outside of information technology, the most substantive investment in dedicated AI companies has gone into financial services.
Case studies
Let me hit you with a few scenarios that crystallize why this makes sense:
Scenario 1: The AI Startup
You’re a London-based AI company. You just raised £10 million. You need to train your first production model. You call AWS : they tell you there’s capacity, but it’s shared with other workloads, and you might experience availability issues during peak times. You call Google Cloud, similar story. You call CoreWeave and they’re at capacity for the next three weeks.
You call Nebius. They’ve got availability. You can start training tomorrow. You’ll pay a bit more than AWS’s list price, but you’re not optimizing for cost at this stage—you’re optimizing for speed. You need to hit your product milestones to raise your Series B. Done deal.
Scenario 2: The Financial Institution
You’re a bank. You’re building a fraud detection system that needs to process transactions in real-time. You need guaranteed compute availability, UK-based infrastructure for data sovereignty, and private clusters for security. The hyperscalers can do this, but their enterprise sales process takes months.
You need something operational in weeks, not months. Nebius specializes in this. They can provision a private cluster, handle the security audits, and get you running faster than the hyperscalers can schedule their kickoff meeting.
Scenario 3: The Research Lab
You’re an AI researcher at Cambridge. You’ve published papers, but now you want to commercialize your research. You need compute to prove your approach scales beyond academic datasets. Your university’s compute allocation is limited. The hyperscalers are expensive for the experimentation phase.
Nebius offers flexible pricing for research-to-commercial transitions. They understand this customer profile because they’ve served it before. They’re willing to work with you on pricing now in exchange for being your provider as you scale.
These aren’t hypothetical. These are the actual customer profiles that exist in the UK market right now. And they’re underserved.
The Counterargument I Keep Wrestling With
Of course everything is not easy. The obvious counterargument is: if this opportunity is so obvious, why isn’t everyone doing it?
Fair question. A few thoughts:
First, building data centers and securing GPU allocations requires capital. A lot of capital. CoreWeave has raised billions, but Nebius can use the stakes in its subsidies to finance without going for debt or dilution.
Second, there’s execution risk. Just because you have GPUs doesn’t mean customers will use them. You need to build a platform, sales motion, customer success organization. You need to actually operate the infrastructure reliably. That’s hard.
Third, and this is the most important one : what happens when the GPU shortage truly ends?
Won’t pricing collapse? Won’t customers migrate back to AWS for the convenience of having compute integrated with the rest of their cloud infrastructure?
Maybe. But I think the specialization moat is deeper than that. Financial firms will still want UK-based, private clusters. Startups will still want providers who understand their specific workflow needs. The hyperscalers are building for everyone—specialist providers can optimize for specific verticals.
And timing matters. If Nebius can build sticky relationships with customers now, while they’re supply-constrained and eager for alternatives, those relationships compound. Your data is there. Your tooling is integrated, andwitching costs accumulate.
What this means for investors
I started writing this to understand Nebius’s UK strategy, not to make an investment case. But I can’t ignore the implications.
If I’m right about this, if the UK really is a structurally attractive market for specialist GPU cloud providers, if Nebius executes well on standing up infrastructure and acquiring customers, if the regulatory environment remains favorable (these are a lot of ifs, I know) then this isn’t just a geographic expansion but rather the 1st push into being the reference cloud for sovereign AI. It’s a blueprint for how to compete with hyperscalers in the AI infrastructure layer.
You cannot really beat hyperscalers globally, but you can beat them in specific markets where you have advantages and where you understand the local customer needs better, or even where regulatory considerations create opportunities for specialized providers.
The risk is execution. Can they actually stand up reliable infrastructure? Can they build a sales organization that converts prospects? Can they deliver a customer experience that makes people want to stay?
I don’t know. But with the UK AI sector attracting £200 million per day in investment, so the question really is supply-side execution.
Conclusion
Coming back to where I started: infrastructure plays are boring until they’re not. And right now we’re at an inflection point where boring infrastructure is actually the smartest bet in AI.
Outside of finX, everyone’s focused on the models and AI products, without thinking too much about the fact that someone needs to provide the compute that makes all of that possible. And in a market where demand is exploding faster than hyperscalers can build capacity, there’s real money to be made being the provider who’s actually available when customers need you.
Nebius’s UK expansion is about positioning themselves at the intersection of several trends that all point in the same direction:
Sovereign AI needed in European countries
Explosive AI investment and startup growth in the UK
Regulatory flexibility post-Brexit that makes the UK more attractive than EU alternatives
Supply constraints for GPU compute that specialist providers can exploit
Enterprise and financial services demand for UK-based, secure infrastructure
First-mover advantage in a market that’s large enough for multiple providers but not yet saturated
Sometimes the boring answer is the right answer.
Disclosure : I own shares in Nebius Group. This article does not constitute financial advice.
CRWV's £1B UK investmant is actualy way smarter than most people realize. The finacial services angle you mentioned is huge becuase those clients stick for compliance reasons not just performace. If they can lock in those relationships befor hyperscalers wake up to it, thats defensible revenue.