AI infrastructure news in the UK is like London buses - nothing for ages, and then, all at once.

Stargate UK is official. The UK is to host the largest European GPU cluster. Google, Microsoft, BlackRock, and a host of others are part of a new UK-US tech deal.

And Nscale?

They’re right in the middle of all of it.

I’m Ben Baldieri. Every week, I break down what’s moving in GPU compute, AI infrastructure, and the data centres that power it all.

Here’s what’s inside this week:

Let’s get into it.

The GPU Audio Companion Issue #65

Want the GPU breakdown without the reading? The Audio Companion does it for you, but only if you’re subscribed. If you can’t see it below, click here to fix that.

Stargate Lands in the UK via Nscale

The UK has its Stargate.

OpenAI, NVIDIA, and Nscale are launching Stargate UK, a sovereign AI infrastructure platform designed to run OpenAI’s flagship models on British soil. The project begins with 8,000 GPUs in Q1 2026, scaling to 31,000 GPUs, and will span multiple sites, including Cobalt Park in the North East, now designated an AI Growth Zone.

For Nscale, this is more than one deal.

It’s a full-stack commitment.

Nscale is also deploying 4,600 GB300s into NVIDIA DGX Cloud, including the new Lepton Marketplace, while rolling out up to 58,640 GPUs across the UK as part of its 300,000-GPU global expansion. On top of that, Nscale’s AI Campus in Loughton will deliver 50MW (scalable to 90MW) and host 23,040 NVIDIA GB300 GPUs by Q1 2027, in partnership with Microsoft, making it the UK’s largest AI supercomputer and the backbone of Microsoft’s Azure services nationwide.

Oh, and Microsoft are also getting involved with Stargate Norway. Because why not?

Why this matters:

  • Stargate UK guarantees OpenAI’s models can run locally under UK jurisdiction, a requirement for regulated industries and public services.

  • From Stargate Norway with Aker ASA (and now Microsoft) to Stargate UK with OpenAI, NVIDIA, and Microsoft, Nscale is now OpenAI’s European anchor partner.

  • The timing aligns with the US-UK Tech Prosperity Deal, under which American tech giants pledged a total of £31B into UK AI infrastructure. Nscale’s role places a UK-headquartered neocloud at the centre of this transatlantic effort. Not bad for a company that only emerged from stealth 18 months ago.

Click here to read the full announcement from Nscale, NVIDIA, OpenAI, and the UK Government.

Carbon3.ai Launches, Nebius Funds Microsoft Deal

Two very different funding stories, one clear signal: capital is pouring into the neoclouds.

Carbon3.ai has launched a UK-based sovereign AI platform, aiming to build a nationwide “AI mesh” of data centres with capacity for more than 100,000 GPUs. Backed by HPE, Vast Data, and WWT, the project pitches itself as a “national grid for AI” powered by off-grid renewables. The £1B growth plan is still light on disclosed investors, but the scale, 30+ UK sites in play, is designed to hard-wire sovereign AI into Britain’s infrastructure map.

Meanwhile, Nebius has gone global with a $4.2B raise split between a Nasdaq share sale and convertible notes.

Proceeds will fund last weeks $17.4B Microsoft contract and accelerate global expansion, from compute and land buys to new data centres. On top of that, Nebius has rolled out the Robotics and Physical AI Awards, dangling $100K in credits and global visibility to startups at the frontier of embodied AI.

Why this matters:

  • Sovereign AI ambitions are shifting from slogans to shovels, and Carbon3.ai is betting big on localisation, renewables, and regulatory-friendly design.

  • While relatively “late” to the proverbial neocloud party, Carbon3.ai’s partnerships with Vast Data, HPE, and WWT, and a 100,000-GPU target are already more than some of the more “established” players have achieved.

  • Nebius shows how scale is banked: Microsoft offtake + Wall Street cash = a financing model mid-tier neoclouds can’t easily replicate. And an execution model to envy.

Click here to read the Carbon3 story from DCD.

Click here to read the Nebius announcement.

CoreWeave Secures $6.3B Backstop from NVIDIA

CoreWeave has landed a $6.3B capacity guarantee from NVIDIA.

This is the second offtake deal the chipmaker has made in as many weeks, and ensures Team Green will buy all of CoreWeave’s unsold compute through 2032. The deal extends their 2023 agreement, giving CoreWeave enviable insulation from demand shocks.

What NVIDIA will be using this much capacity for is anyone’s guess. New product releases geared towards physical AI, and Brookfield’s new partnership with Figure could be a hint at the direction of travel. As with agentic AI, however, the path to widespread adoption remains to be seen.

Why this matters:

  • First Lambda, and now this, shows NVIDIA is doubling down on select neoclouds. The question is why?

  • With highly leveraged balance sheets across the sector, deals like this shore up the biggest players (ie those with the largest amount of hardware that could hit the secondary market if things go sideways), reducing the risk of an implosion and subsequent supply shock (bad for selling new generations of hardware).

  • NVIDIA’s guarantee effectively assures CoreWeave’s utilisation for the next seven years. Given the strength of NVIDIA’s balance sheet, and the fact that the terms of most financing in this space are dependent upon the creditworthiness of the offtaker, the neocloud likely now has the flexibility needed to refinance its heavy debt burden.

Click here to read the full story on Reuters.

Physics AI Gets Boost with Luminary Cloud’s $72m Raise

Luminary Cloud has secured a $72m Series B round led by N47, with participation from Sutter Hill Ventures and NVentures.

While most neoclouds chase general-purpose AI, Luminary is carving out a vertical of its own: Physics AI. Its platform uses synthetic data to train models that predict the performance of real-world products (think cars, aircraft, and electronics) in near real-time. The raise will fund the expansion of Luminary’s Physics AI Model Factory, its MCP Server for faster model development, and an AI-assisted Notebook for design workflows. Early customers include Honda, Otto Aerospace, Joby Aviation, Piper Aircraft, and Trek Bikes.

Why this matters:

  • While others compete for the same training and inference workloads, Luminary is anchoring itself in industrial design and engineering, a market with fewer rivals and deeper moats.

  • The deeper into a given niche a neocloud can go, the less competition it’ll have, and the market for physics AI-tailored neoclouds is currently a market of one.

  • Product engineering is a trillion-dollar market. Physics AI could shorten design cycles and shift manufacturing competitiveness, opening new demand for AI infrastructure.

Click here to read the full announcement from Luminary Cloud.

NVIDIA Backs Intel as Huawei Ascends

NVIDIA is investing $5B into Intel, taking a ~4% stake and agreeing to co-develop multiple generations of PC and data centre chips.

The move follows the US government’s $8.9B equity injection, cementing Intel’s role as Washington’s chosen turnaround bet. The partnership will see Intel CPUs paired with NVIDIA GPUs using proprietary high-speed links - a potential blow to AMD as it looks to scale the Instinct accelerator market penetration. In the broader context of an inconsistent export regime, the lasting market impact remains unclear.

NVIDIA’s H20 and subsequent B20 chips were first banned from export, then conditionally allowed under a US profit-share model.

Beijing has, in response, prohibited local firms from using NVIDIA chips, doubling down on local alternatives, with Huawei seizing the moment.

At its Shanghai event, the company unveiled an aggressive Ascend roadmap: the 950 series in 2026, 960 in 2027, and 970 in 2028. These chips promise up to 8 PFLOPS FP4 performance, 4 TB/s memory access, and 16 PB/s interconnect bandwidth. Huawei’s Atlas 950 SuperPoD is pitched as six times faster than NVIDIA’s planned NVL144, and 62x higher on interconnect, built to anchor China’s burgeoning domestic AI stack.

Why this matters:

  • NVIDIA’s Intel stake embeds it firmly in US industrial strategy, even as its China exposure weakens.

  • Huawei is doubling down on Ascend as the domestic and export-friendly alternative, tuned for Chinese open-source models and local regulation.

  • Countries outside the US-China bloc face fewer neutral options. Deals with NVIDIA or Huawei will increasingly carry political strings, shaping who controls AI infrastructure in Africa, Latin America, and South Asia.

Click here to read the full story from Reuters.

Google Opens £5B UK Data Centre

Google has opened a new data centre in Waltham Cross, Hertfordshire, as part of a £5B investment package over the next two years.

The facility will handle workloads across Cloud, Search, and Maps, while underpinning DeepMind’s research in science and healthcare. To secure clean power, Google has signed an agreement with Shell to manage its UK carbon-free portfolio and deploy new battery systems for storage and grid balancing. The spend covers capex, R&D, and engineering tied to both Google Cloud and DeepMind, with the site projected to support 8,250 jobs annually.

Why this matters:

  • A £5B pledge cements Google’s role in the UK’s AI economy at a time when Washington’s tech agenda is pulling capital back to the US.

  • Grid connections and power are a major bottleneck in the UK AI infrastructure market, so the Shell partnership (and renewable focus) could signal the beginning of a new phase of market development, with oil majors right at the heart of it.

  • With Chancellor Rachel Reeves on-site for the opening, this is as much about AI jobs, economic growth, and scoring political wins as it is about compute.

Click here to read the full announcement from Google.

BlackRock Commits £500M to UK Data Centres

BlackRock has committed up to £500M ($678M) to UK data centres through a new joint venture with Gravity Edge.

The venture, Digital Gravity Partners, will focus on acquiring existing facilities with underutilised capacity and, crucially, secured power. The first site, in West London, will seed the platform with an initial £100M investment from BlackRock’s €1.2B Europe Property Fund VI, raised in July.

Unlike the hyperscale mega-campuses that dominate headlines, Digital Gravity is targeting smaller enterprise-oriented sites. These facilities require less capex to reposition, can be upgraded faster, and offer a broader pool of exit buyers, including infrastructure and real estate funds.

UK conditions make this strategy attractive. Power prices are among the highest in Europe, and grid access is constrained, raising barriers for greenfield development. Acquiring brownfield sites with existing connections avoids long planning and grid-queue delays while locking in scarce megawatts at a discount.

Why this matters:

  • Rather than chase hyperscalers with multi-billion-dollar builds, BlackRock is carving out a niche in mid-sized enterprise sites with faster turnaround and clearer exit liquidity.

  • With the UK grid under strain, power-secured assets carry a premium, meaning BlackRock’s bet is on scarcity economics.

  • Wall Street capital is moving beyond 500MW+ campuses into opportunistic plays that monetise overlooked, stranded, or underutilised assets.

Click here to read the full story on Bloomberg.

The Rundown

Where to begin.

The AI infrastructure buildout has had a slow start in the UK, largely due to the country's power issues. Problem one is there isn’t enough of it available, and two is it’s the most expensive in the world. This has led many UK-based neoclouds to seek data centre capacity in Scandinavia, where power is cheaper and margins are thicker. Ideal for startups where every $ of OpEx counts.

So, what’s changed?

Market maturity.

The bigger neoclouds are no longer startups (although perhaps not from a creditworthiness perspective. That’s reflected in the scale of the deals and the size of the counterparts. Organisations are clearly maturing, and the winners of the past 24 months of cutthroat competition are making their presence felt. And credit where credit is due, contracts and offtake like this are incredibly difficult to land, because NVIDIA, Microsoft, and OpenAI aren’t going to sign with just anyone.

But these moves also aren’t charity, because none of these companies does anything unless there’s significant upside in it for them as well.

This week, therefore, means NVIDIA, OpenAI, Microsoft, BlackRock, Google, and Washington keep the UK firmly locked in as a key partner in the global AI race.

Jensen has cemented his position as a kingmaker from the top to the bottom of the broader semiconductor geopolitical neocloud ecosystem. Intel now has a massive lifeline that few could have predicted ten years ago. And all perfectly timed to happen just as Huawei announces what could develop into a meaningful challenge to US dominance in the global south.

Funny that.

See you next week.

Reply

or to participate