SWI Compute announces European AI factory plans with NVIDIA systems

SWI Group announced the launch of SWI Compute on October 26, positioning secured electricity as the foundation for a European AI cloud business. The company says it has ordered approximately $1 billion of NVIDIA systems. Its parent's 3.6 GW of secured power represents access to electricity, rather than AI capacity already operating.

The UK plan begins with 30 MW, nearly 7,000 Rubin GPUs and 3,500 Vera CPUs in Vera Rubin NVL72 racks, expected online by the end of 2027. The October announcement puts the eventual campus at 530 MW and targets 2.3 GW of AI factory capacity by 2029. Both remain future development plans.

A separate initial deployment specifies Blackwell Ultra, with systems partners selected to deliver more than 8,000 GPUs; it should not be combined with the UK Rubin phase as one delivered fleet. An unnamed US inference platform has contracted capacity. SWI plans to use NVIDIA DSX for future facilities. The order figure does not establish what has already been paid or delivered.

The launch follows an earlier change in the group's route into computing. On August 25, SWI said it would develop an internal AI cloud platform for enterprises, researchers and developers. It also announced that a previously proposed majority investment in Polarise would not proceed. Instead, SWI would finance Polarise's expansion while the businesses remained separate. That earlier statement established the in-house strategy now taking shape under the SWI Compute name, linking land and power assets with services sold to computing customers.

Financing had also advanced before the launch. In its October 12 announcement, SWI disclosed an €800 million capital increase involving institutional investors and new ordinary shares issued in several series. The group said proceeds would support power infrastructure, computing deployments and its development pipeline across Europe and the United States. This was a group-level funding announcement covering a wider expansion program, rather than a disclosed budget solely for the UK factory.

An AI factory is a data center organized around training or running AI models, with processors, networking and facility systems designed together. NVIDIA's DSX product documentation describes software, reference designs and simulation tools spanning that infrastructure. DSX Sim models facilities and computing before construction; DSX MaxLPS coordinates workload placement and power across GPUs and racks. Those functions help explain the engineering approach SWI is adopting, without establishing how a particular future site will perform.

NVIDIA's separate October 21 European infrastructure announcement provides more detail about operation. DSX Flex coordinates computing demand with grid conditions, while the DSX OS software portfolio includes network configuration, rack management and cluster readiness checks. The new management components were described as developer previews. NVIDIA also explains that closed-loop liquid cooling can move heat to outdoor dry coolers without evaporating cooling water, while local climate can still require supplemental mechanical cooling during the hottest periods.

The platform's design work predates this week's announcements. NVIDIA introduced the Omniverse DSX blueprint in October 2025, describing digital twins that combine building layouts with electrical and thermal models. Engineers can use those models to examine designs and anticipate failures before construction. Its original account also describes factory-built modules and using the digital twin after commissioning to monitor and optimize operations. That lifecycle approach connects facility planning with the ongoing work of keeping installed AI systems productive.