AI Agent UpdateSWI Compute emerges from stealth to turn gigawatts of power into Nvidia AI factories
SWI Compute spent six years securing 3.6 gigawatts of power across Europe and the US before selling a single watt of compute — now it is turning that power into Nvidia AI factories, starting in the UK.
The News
SWI Compute emerged from stealth on October 26 with a simple pitch: it spent six years securing power first, and only now is it selling compute. The company, part of SWI Group, says it has lined up 3.6 gigawatts (GW) of grid and power capacity across 24 sites in Europe and the US, and is now converting that position into AI factories built on Nvidia systems. A gigawatt is a billion watts — roughly the output of a large power station — and in the AI era, power has become the scarcest ingredient: you cannot run thousands of AI chips without an enormous, reliable electricity supply.
SWI Compute says it became a Preferred NVIDIA Cloud Partner in August, has ordered about $1 billion of Nvidia systems, and has signed its first customers. Its first factory, in the UK, starts at 30 megawatts (MW) — about 7,000 Rubin GPUs and 3,500 Vera CPUs arranged in Vera Rubin NVL72 racks, due by the end of 2027. Rubin is Nvidia's next-generation AI chip platform, and an NVL72 rack is a cabinet packing 72 of these chips together so they can work as one machine. The UK site is phase one of a 530 MW campus, with the company aiming for 2.3 GW by 2029.
The announcement comes from the company itself — a press release via PR Newswire — so it should be read as SWI's plan and claims, not independently verified construction progress. What makes it stand out is the sequencing: most AI infrastructure firms announce compute first and hunt for power later, often discovering that grid connections take years. SWI says it did the slow part first, spending six years on power, which is why it can now promise factories on a timetable. Whether those timetables hold will be proven by steel in the ground, not by press releases — but the order of operations is the story.
Power first, compute second — the slow part was done before the launch, not after it.
Why It Matters
The first reason this matters is the power bottleneck. Across Europe and North America, data-centre projects are queuing for grid connections that can take years, and some have been cancelled or delayed for lack of power. A company arriving with 3.6 GW already secured is therefore selling something rarer than GPUs: the right to switch the machines on. That is why SWI frames power, not chips, as its core asset — and why its 24-site footprint matters more than any single factory announcement.
The second reason is the Nvidia link. Preferred partner status, a $1 billion systems order and first customers signed give the plan more weight than a typical stealth launch: Nvidia does not hand out preferred status casually, and a billion-dollar order means real hardware is earmarked. For Nvidia (NVDA) — named here as a public read-through, since SWI Group itself trades in Amsterdam as SWICH and Nvidia is the supplier — every new AI factory is another customer for its chips, racks and software. The read-through is straightforward: more factories, more Nvidia systems sold.
The third reason is the European angle. Much of the AI factory race has been a US story, with power and land constraints pushing some projects abroad. A UK-first, Europe-led build-out, backed by gigawatts secured across the continent, suggests Europe is becoming a serious site for AI infrastructure rather than just a customer of it. If SWI delivers its 30 MW start by end-2027 and scales toward 2.3 GW by 2029, it would be one of the larger European AI infrastructure stories — but those are targets, and the gap between a launch announcement and a working factory is where most such plans are tested.
Analyst's Take
The verdict is Bullish because the pieces fit in the right order: power secured before compute sold, a preferred Nvidia partnership, a billion-dollar systems order, first customers signed, and a concrete first milestone — 30 MW in the UK by end-2027. The risks are real — this is a company announcement, not a completed build, and the 2.3 GW target for 2029 is ambitious — but the sequencing is exactly what the power-constrained AI infrastructure market rewards. For Nvidia, it is another demand signal; for Europe, it is a sign the AI factory race is no longer America-only. Readers' poll: can SWI's power-first approach deliver its UK factory by end-2027?
By The Numbers
FAQ
What is a gigawatt, and why does power matter so much for AI?
A gigawatt (GW) is a billion watts — roughly the output of a large power station. AI data centres pack thousands of power-hungry chips together, so a single factory can need as much electricity as a small city. Across Europe and North America, grid connections for such projects can take years, which makes secured power one of the scarcest ingredients in the AI build-out.
What is a Vera Rubin NVL72 rack?
Vera Rubin is Nvidia's next-generation AI chip platform, pairing Rubin GPUs (the processors that do the AI calculations) with Vera CPUs (the general-purpose processors that coordinate them). An NVL72 rack is a cabinet packing 72 of these chips together so they can work as one machine. SWI's first UK factory will hold about 7,000 Rubin GPUs and 3,500 Vera CPUs in such racks.
Is SWI Group a public company, and why is Nvidia's ticker here?
SWI Group trades on Euronext Amsterdam under the ticker SWICH. Nvidia (NVDA) is named here as a public read-through: Nvidia is the supplier of the systems SWI is buying, so every SWI factory is another customer for Nvidia's chips, racks and software. NVDA is the listed name through which readers can follow the demand side of this story.
Source: PR Newswire