UK AI Growth Zones 2026: GPU Cloud & Data Residency

The UK is not the EU, and treating "European data residency" as one bucket is the fastest way to get a UK compliance review wrong. Since Brexit, the UK runs UK GDPR as a separate legal regime from EU GDPR, with its own regulator, its own transfer mechanisms, and a deliberately lighter AI policy than Brussels. At the same time, the UK just landed the largest sovereign-GPU commitment in its history: 120,000 Blackwell Ultra GPUs and up to £11 billion in investment, led by Nscale, Nebius, and CoreWeave under the government's AI Growth Zones program.

Most of that capacity is not live yet. This post maps what's actually operational versus announced, walks through the UK GDPR/EU GDPR split in concrete terms, and gives UK teams a real path to renting H100, H200, or B200 GPUs today rather than waiting on Growth Zone buildout or routing everything through a US hyperscaler region. For the equivalent picture across the Channel, see our GPU cloud providers Europe guide, and for how a different non-EU market handles the same trade-offs, see our GPU cloud providers Australia guide.

The UK's AI Growth Zones and What's Actually Being Built

NVIDIA, Nscale, CoreWeave, and Microsoft are building the UK's next generation of AI infrastructure: 120,000 Blackwell Ultra GPUs, backed by up to £11 billion in investment, targeted to be operational by the end of 2026, according to NVIDIA's own announcement. That makes it the largest single AI infrastructure rollout in the country's history.

"The United Kingdom is building the infrastructure for the AI industrial revolution, advancing science, transforming industries and creating new economic opportunities," said Jensen Huang, NVIDIA's CEO, in that announcement. Sam Altman added: "Stargate U.K. builds on this foundation to help accelerate scientific breakthroughs, improve productivity and drive economic growth."

The Five Designated AI Growth Zones

The UK government has designated five AI Growth Zones so far, sites picked for access to at least 500MW of power and streamlined planning support:

Collectively, the first cohort has secured £28.2 billion in private investment and is projected to create more than 15,000 jobs, per a tracker of the program. The North Lanarkshire zone alone, the newest of the five, is expected to attract $11.2 billion in private investment and generate over 3,400 jobs, according to US government trade analysis of the program.

Being "designated" is a planning and power-access status, not a live data center. Of the five, North Lanarkshire is the only zone with a named, funded operator already building on-site: DataVita, in partnership with CoreWeave.

Nscale's 60,000-GPU Buildout and Stargate UK

Nscale, a UK-based neocloud, has earmarked 60,000 NVIDIA GPUs for the UK as part of a 300,000-GPU global buildout. It's also the anchor for Stargate UK, a joint infrastructure platform with OpenAI and NVIDIA focused on sovereign workloads.

Two things are actually under contract here, and they're easy to conflate:

OpenAI's compute offtake. OpenAI is exploring uptake of up to 8,000 GPUs starting in Q1 2026, with room to scale to 31,000, according to Nscale's own press release on the deal. Part of this runs at Cobalt Park in the North East AI Growth Zone.

The Microsoft-Nscale Loughton deal. Separately, Nscale and Microsoft have committed to build what they're calling the UK's most powerful supercomputer, at a Loughton, Essex campus for Azure services: 23,040 NVIDIA GB300 (Grace Blackwell Ultra) GPUs, on 50MW of capacity scalable to 90MW, arriving Q1 2027.

Neither of these is live yet. The 8,000-GPU OpenAI tranche is an exploratory commitment for Q1 2026, and Loughton's GPUs don't land until 2027. If your workload needs UK-region Stargate capacity today, it doesn't exist yet at any meaningful scale.

Nebius, CoreWeave, and Microsoft's Parallel Bets

Three more sovereign bets are running alongside Nscale, at different stages of readiness.

Nebius has the most GPU hardware physically racked in the UK today. Its cluster at Ark Data Centres' Longcross Park campus in Surrey deployed 4,000 NVIDIA HGX B300 (Blackwell Ultra) GPUs in its first phase, inside a 54MW footprint across two data halls, with the full project set to scale to around 9,000 Blackwell Ultra GPUs, per reporting on the Ark-Nebius deal. Ark itself is putting £7.5 billion into UK AI-ready capacity as part of the arrangement. "We're pleased to be providing compute infrastructure that will support future innovation by British businesses, researchers and the public sector," said Nebius founder and CEO Arkady Volozh when the UK expansion was first announced.

CoreWeave committed £1.5 billion in new UK investment, bringing its total to £2.5 billion, to deploy Grace Blackwell Ultra and GB300 GPUs at DataVita's data center in North Lanarkshire, Scotland, per CoreWeave's own announcement. The site runs on renewable energy with closed-loop cooling, and it's the anchor deployment behind the North Lanarkshire AI Growth Zone.

Microsoft shows up twice in this list, both as the Azure customer behind the Nscale Loughton buildout and as a broader UK Azure region operator independent of the Growth Zones program.

Of these, Nebius's Longcross cluster is the only one with GPUs confirmed operational (Q4 2025). CoreWeave's Scotland site and the Nscale-Microsoft Loughton campus are under construction or in early deployment through 2026 and 2027. For a wider look at how these same neocloud names are positioned across other regions, see our Nscale alternatives comparison and our breakdown of Nscale's Anyscale acquisition.

UK GDPR and Data Residency: How It Differs From the EU

Storing data in a UK data center satisfies UK GDPR. It does not automatically satisfy EU GDPR, and the reverse is also true, even though the two frameworks started as identical text.

Two Separate Regimes Since Brexit

UK GDPR and EU GDPR were the same law on 31 January 2020, the day the UK left the EU. Since then they've diverged: separate enforcement bodies (the Information Commissioner's Office in the UK versus the European Data Protection Board for the EU), separate legislative amendment paths, and separate adequacy mechanisms, according to an overview of post-Brexit UK data residency rules. The UK's Data (Use and Access) Act has already amended the UK's domestic framework in ways that don't automatically mirror EU changes.

For cross-border transfers, the UK uses its own instruments rather than the EU's Standard Contractual Clauses directly: the International Data Transfer Agreement (IDTA), or the UK Addendum bolted onto EU SCCs for entities that need both. Treating UK and EU residency as interchangeable is a compliance risk, not a shortcut. A vendor with an EU GDPR-compliant DPA is not automatically UK GDPR-compliant, and you should ask for UK-specific evidence rather than assume equivalence.

The EU-UK Adequacy Decision and What It Actually Covers

On 19 December 2025, the European Commission renewed its UK adequacy decisions, both under GDPR and the Law Enforcement Directive, extending them through 27 December 2031, a six-year window, per legal analysis of the renewal. That followed an interim extension earlier in 2025 while the Commission assessed how the UK's Data (Use and Access) Act might affect the UK's data protection standards.

Adequacy means personal data can flow from the EU to the UK without extra transfer mechanisms like SCCs. It does not mean the two regimes are the same regime. It's a bridge for data flow, not a merger of the two legal frameworks, and it's reviewed periodically rather than granted permanently, so it carries its own renewal risk on a multi-year horizon. Our sovereign AI buyer's guide to data residency covers the broader mechanics of matching compute placement to a given jurisdiction's rules.

UK GDPR vs EU AI Act: A Lighter-Touch Approach

Where the UK diverges most sharply from Brussels is AI-specific regulation. The EU AI Act is a risk-tiered law: it classifies AI systems by risk level, mandates conformity assessments for high-risk systems, and prohibits some uses outright. The UK has no equivalent statute. Instead, it runs a principles-based, sector-led model, where existing regulators apply general AI principles within their own remit rather than a single cross-sector law policing AI systems directly, according to a comparison of UK and EU AI governance.

Practically, this means a UK-only AI deployment carries a lighter formal compliance checklist than the same system serving EU users. It does not mean no obligations: UK GDPR's Article 22 on automated decision-making still applies, and the ICO has shown it will interpret existing law flexibly rather than leave a gap. For the EU side of that comparison, our EU AI Act compliance guide for GPU cloud covers the conformity-assessment obligations a UK-only deployment doesn't have to clear.

Renting GPUs in the UK Before AI Growth Zones Capacity Lands

The AI Growth Zones headline numbers are all future capacity. If you need H100, H200, or B200 access in the UK this quarter, here's what's actually rentable right now.

Hyperscaler UK Regions: What's Actually On-Demand

AWS runs P5 (H100 SXM5) instances in its London region (eu-west-2) today, available on-demand, through Capacity Blocks for ML, and occasionally on spot, though spot P5 capacity is scarce given persistent on-demand and reservation demand. London is one of AWS's stronger EU-adjacent regions for accelerated computing breadth, spanning P4d and P5 families, though exact availability depends on your account's service quota and should be checked directly in-console rather than assumed. Azure and GCP both operate UK regions as well, but their newest GPU generations (H200, B200) land in UK-specific regions later than US-East, the same pattern documented in our GPU shortage 2026 analysis for EU regions generally.

UK Neocloud and Sovereign Options vs Global Marketplaces

Nscale, Nebius, and CoreWeave are all UK-present today, but none of them sell self-serve, instant-provision GPU capacity the way a marketplace does. Nscale in particular has no published self-serve pricing; teams researching it typically end up comparing it against providers that do quote a rate directly, which our Nscale alternatives roundup covers in detail. Their sovereign deployments are built for large, contracted allocations tied to Growth Zone capacity, not a single researcher spinning up one H100 for an afternoon.

That's the gap a global GPU marketplace like Spheron fills: instant provisioning, per-minute billing, and access to H100, H200, and B200 capacity through 5+ providers, without a sales process or a wait for a specific UK data center's allocation to open up. It's not a UK-sovereign platform itself, so for workloads with a hard UK-only or CLOUD Act-free placement requirement, the hyperscaler UK regions or a UK-incorporated provider are the right call. For training, batch inference, and development workloads where regional placement isn't a strict regulatory requirement, Spheron is the faster and cheaper path.

Pricing: UK/Hyperscaler Rates vs Spheron

AWS's P5.48xlarge, the standard H100 SXM5 instance available in London, lists at roughly $6.88/hr per GPU on-demand after AWS's 2025 rate cut, a figure our AWS H100 pricing breakdown covers across all P5 variants and hidden costs. That's before EBS, EFA data transfer, and egress, none of which show up on the headline instance rate.

Pricing fluctuates based on GPU availability. The prices above are based on 20 Aug 2026 and may have changed. Check current GPU pricing → for live rates.

The gap holds even before you account for AWS's line-item extras. Spheron's H100 SXM5 on-demand rate is roughly half AWS London's list price for the same silicon, and spot brings it closer to a third. For UK teams building on a global marketplace, H100 SXM5 rental and H200 GPU pricing are both live today rather than tied to a 2026 or 2027 Growth Zone timeline.

Decision Tree: Which Setup for Which UK Workload

Use this to match your actual constraint, not the headline you saw, to the right infrastructure:

Strict UK-only data residency, no CLOUD Act exposure tolerance. Use a UK-incorporated provider or a hyperscaler UK region with a signed UK Addendum/IDTA in place. AWS's London P5 instances satisfy this for H100 today; H200 and B200 UK-region on-demand capacity is still catching up through 2026.

EU users, UK-only compute won't satisfy the requirement. UK adequacy lets EU personal data flow to the UK, but it doesn't turn UK infrastructure into EU-resident infrastructure for regulatory purposes. If your obligation is specifically EU data residency, use an EU-region provider, covered in our Europe GPU cloud guide, not a UK one.

Training on non-personal or anonymized data, cost is the constraint. Regional placement carries minimal compliance weight here. B200 GPU availability on Spheron at spot pricing, with checkpoints written to UK or EU object storage, is the cheapest path to Blackwell-class hardware without waiting on a Growth Zone allocation.

Waiting on Stargate UK, Loughton, or another named sovereign deployment for production capacity. Most of the 120,000-GPU commitment lands through the second half of 2026 and into 2027. Plan your current workload against what's live now, via Spheron or a hyperscaler UK region, and treat sovereign capacity as a future migration rather than a near-term dependency. Spheron's documentation on dedicated vs spot instance types covers the trade-offs for a workload that might move once UK-sovereign capacity is actually available.

The UK's sovereign GPU buildout is real but mostly not live yet. Spheron gives UK teams H100, H200, and B200 access through 5+ providers today, with per-minute billing and no wait on Growth Zone capacity.

H200 on Spheron → | Check B200 availability → | View all GPU pricing →