AI News
SWI Compute Emerges From Stealth With 3.6 GW of Power for NVIDIA AI Factories Across Europe
SWI Compute has emerged from stealth with one of the largest power positions in European AI infrastructure: 3.6 GW of capacity secured over six years across 24 sites in Europe and the United States. The company is a new unit of SWI Group, which is listed on Euronext Amsterdam, and it plans to turn that power into NVIDIA AI factories.
One dating note first, because it matters for readers following this story. SWI Compute emerged from stealth on October 26, 2026, when its launch was announced. Coverage on November 2, including the widely shared Intelligent CIO report, described that announcement rather than a new November launch. This article uses the October 26 date for the launch itself.
The plan is to build 2.3 GW of NVIDIA AI factory capacity by 2029. Five European campuses are named — in Ireland, the UK, Denmark, Spain and Italy — and one campus is already leased to a major hyperscaler, which has not been named. The first customers signed include a US-based AI inference provider, also unnamed.
The flagship is the UK site. It starts at 30 MW with close to 7,000 Rubin GPUs and 3,500 Vera CPUs in Vera Rubin NVL72 racks, and is due online by the end of 2027. The campus is planned to grow to 530 MW. For scale, 7,000 of NVIDIA's latest GPUs in a single initial deployment is a serious cluster, and a 530 MW campus would be among the largest AI sites in Europe.
The NVIDIA relationship runs deeper than a purchase order. SWI Compute became an NVIDIA Preferred Cloud Partner in August, has ordered about $1 billion of NVIDIA systems, and builds to NVIDIA reference architectures under the DSX AI Factory platform. The company claims that platform can enable up to 40% more GPU capacity within the same power budget — a company claim, not an independently verified figure. The first deployment runs Blackwell Ultra systems, with Vera CPU racks to follow, connected by Quantum-X800 InfiniBand and Spectrum-X Ethernet networking. Delta Computers is the systems integrator, with more than 8,000 GPUs to its name.
The real story here is power, not chips. Anyone with money can order GPUs; almost nobody can plug gigawatts of them into the grid quickly. Grid connections, permits and substations take years, which is why SWI Group's six-year effort to secure 3.6 GW is the asset that matters. Europe's AI factory race is, in large part, a race for megawatts, and SWI Compute is betting that its head start is worth more than any single hardware generation.
There are caveats to keep in view. The 2.3 GW is a target for 2029, not operating capacity, and the UK site's precise location has not been pinned down in the release, though coverage points to Cambridge. One analysis also noted that the company's Cambridge page lists 500 MW where the release says 530 MW — a discrepancy that has not been reconciled. The hyperscaler tenant and the inference customer are unnamed, so the demand side of the story is still unverified.
Some background on the hardware helps explain the numbers. Vera Rubin is NVIDIA's platform after Blackwell, pairing Rubin GPUs with Vera CPUs in NVL72 racks — 72 GPUs linked as a single system. A deployment of close to 7,000 Rubin GPUs with 3,500 Vera CPUs is therefore not a room of servers but a tightly coupled supercomputing fabric, built for training and large-scale inference alike. The networking matters as much as the chips: Quantum-X800 InfiniBand and Spectrum-X Ethernet are what keep thousands of GPUs fed with data instead of idling.
Europe is the right place to watch this race. The continent has lagged the United States in AI compute, and both the EU and national governments have been pushing sovereign AI capacity — including the AI factories initiative — precisely to close that gap. SWI Compute's pitch is that it can offer what governments and hyperscalers both want: large, ready-powered sites built to NVIDIA's own designs, without waiting years for grid connections. Whether that pitch converts into signed tenants at scale is the question the next two years will answer.
What to watch: whether the UK site hits its end-of-2027 online date, whether the hyperscaler tenant is named, and whether financing and grid milestones keep pace with the 2029 target. If they do, SWI Compute goes from a stealth launch to one of Europe's most important AI landlords. If they slip, the 3.6 GW headline will look like potential rather than progress.