SWI Compute has emerged from stealth to develop and operate NVIDIA-powered AI factories, starting with a planned 30-megawatt deployment in the United Kingdom. Backed by 3.6 gigawatts of power capacity secured by its parent, SWI Group, the new AI infrastructure company plans to supply dedicated and on-demand computing capacity to AI startups, research laboratories and enterprises. The launch includes approximately $1 billion in NVIDIA hardware orders and an expansion strategy targeting 2.3 gigawatts of AI factory capacity by 2029, positioning SWI Compute to compete in Europe’s growing market for high-performance AI infrastructure.
SWI Compute Targets Europe’s AI Infrastructure Gap
AI infrastructure availability is becoming a strategic concern for companies developing and deploying increasingly demanding AI models. SWI Compute, a new venture backed by Euronext Amsterdam-listed SWI Group, is entering this market with plans to build and operate NVIDIA-based AI factories, beginning in Europe.
The company says its parent has secured 3.6 gigawatts of power capacity across Europe and the United States over six years. SWI Compute intends to turn that energy foundation into computing infrastructure for AI-native businesses, research organizations and enterprises that need access to advanced GPUs without building their own data centers.
The company became a Preferred NVIDIA Cloud Partner in August and has ordered approximately $1 billion in NVIDIA systems. It also says it has finalized contracts with its first customers, including a US-based AI inference platform that has contracted computing capacity.
The launch reflects a broader infrastructure challenge: securing electricity, grid connections, cooling and advanced computing equipment can be as important as choosing the AI models an organization wants to run.
A 30 MW UK deployment anchors the expansion
SWI Compute’s first large-scale AI factory is planned for the United Kingdom, beginning with 30 megawatts of capacity. The company expects the initial facility to come online by the end of 2027.
According to the announcement, the first phase is associated with approximately 7,000 GPUs and 3,500 Vera CPUs in NVIDIA Vera Rubin NVL72 rack configurations. The wider UK campus is planned to expand to 530 MW, forming part of a strategy to deploy 2.3 GW of NVIDIA AI factory capacity by 2029.
SWI Group says its power portfolio spans 24 sites, with five European campuses located in Ireland, the UK, Denmark, Spain and Italy. One site has already been leased to a major hyperscaler, according to the company.
The distinction between power capacity and operational computing capacity is important. The 3.6 GW figure represents the group’s reported power capacity, not 3.6 GW of AI computing already running. Similarly, the 530 MW UK campus and 2.3 GW deployment target are future development plans rather than currently available capacity.
The company’s expansion will depend on completing construction, securing equipment, integrating cooling and networking systems, and bringing facilities online according to schedule.
NVIDIA infrastructure becomes the foundation
SWI Compute plans to build its facilities around NVIDIA reference architectures, integrating processors, networking, storage, power and cooling into coordinated systems.
The company says its deployments will use NVIDIA Quantum-X800 InfiniBand and NVIDIA Spectrum-X Ethernet networking. It has selected Delta Computers as a systems integrator, alongside storage partners from the broader NVIDIA OEM and original design manufacturer ecosystem.
The announcement describes a phased hardware strategy involving NVIDIA Blackwell Ultra systems and future Vera Rubin-based infrastructure. It also says new sites from 2027, starting with the UK campus, will follow the NVIDIA Vera Rubin DSX AI Factory reference design.
The hardware roadmap needs clarification before publication: the announcement describes the initial UK deployment both in terms of Vera Rubin NVL72 configurations and an initial Blackwell Ultra deployment, with Vera CPU racks to follow. These descriptions do not fully establish which systems will be installed in each phase.
NVIDIA’s DSX AI Factory Platform is designed to coordinate computing, networking, power and cooling as an integrated infrastructure system. SWI Compute says this approach could enable up to 40% more GPU capacity within the same power budget under the Vera Rubin reference design. That figure is a vendor-stated capability, not an independently verified result from SWI Compute’s facilities.
A cloud infrastructure market shaped by power constraints
The business model targets organizations that need substantial AI compute but may struggle to secure long-term capacity from established cloud providers. AI labs and model developers need infrastructure for pretraining and fine-tuning, while inference providers require reliable computing capacity to serve model outputs to users.
SWI Compute plans to offer both dedicated and on-demand access. Its focus on European deployments could appeal to customers seeking infrastructure closer to their users and data, including organizations with European jurisdictional or data-governance requirements.
The company will nevertheless compete in a market shaped by established cloud platforms, specialist GPU cloud providers and large technology companies investing in their own data centers. Microsoft’s Azure, Amazon Web Services and Google Cloud provide extensive AI infrastructure and managed services, while NVIDIA’s accelerated computing ecosystem supports a wide range of cloud operators.
SWI Compute’s proposed differentiation is its access to power resources secured by SWI Group before the launch of the operating business. If those resources can be converted into functioning AI factories on schedule, the company could reduce one of the constraints facing its expansion. Power access alone, however, does not guarantee competitive pricing, efficient utilization or reliable service.
Turning AI factories into long-lived assets
SWI Compute argues that infrastructure economics depend on how many tokens a facility can generate per megawatt, how long the systems remain productive and how many workloads they can support.
The company intends to serve multiple stages of AI development, including model training, agentic inference, simulation, data processing and scientific computing. Continuous software optimization is expected to help existing systems remain useful as hardware and workloads evolve.
That flexibility will matter as AI demand changes. Operators must balance utilization against electricity costs, hardware depreciation, cooling requirements and the pace of new chip releases.
SWI Group founder and CEO Max-Hervé George described SWI Compute as the final component of a vertically integrated digital infrastructure business extending from grid connections to GPUs. NVIDIA executive Nico Caprez said the company’s planned infrastructure could support Europe’s AI ecosystem, including startups, laboratories, universities and enterprises.
The immediate test is execution. SWI Compute has announced substantial hardware orders, an initial customer contract and an ambitious deployment schedule, but its first UK facility is still planned for the future. Delivering operational capacity on time, at competitive cost and with dependable performance will determine whether its power-first strategy translates into a durable position in the AI cloud market.
4. Market Landscape
AI infrastructure is increasingly constrained by the combination of electricity availability, data-center construction, advanced processors and high-speed networking. SWI Compute is entering this market with a strategy that starts with power capacity and develops dedicated NVIDIA-based facilities around it.
Its potential customers include AI startups, inference platforms, research labs and enterprises seeking alternatives to building proprietary infrastructure or relying exclusively on hyperscale cloud providers.
The competitive environment includes AWS, Microsoft Azure, Google Cloud and specialist GPU cloud operators. SWI Compute’s opportunity rests on converting its reported power portfolio into operational capacity, while offering customers suitable access models, competitive economics and infrastructure aligned with European requirements.
The company’s announced targets remain subject to delivery. Its planned 30 MW UK deployment, 530 MW campus and 2.3 GW capacity goal should be distinguished from live, customer-accessible computing infrastructure.
Top Insights
- Power-led strategy: SWI Compute is using SWI Group’s reported 3.6 GW power portfolio to develop dedicated AI infrastructure across Europe and the United States.
- UK launch planned: The first 30 MW UK AI factory is expected to become operational by the end of 2027.
- NVIDIA ecosystem: Approximately $1 billion in hardware orders and a Preferred NVIDIA Cloud Partner relationship underpin the company’s infrastructure plans.
- European customer focus: The company targets AI startups, research labs and enterprises seeking dedicated or on-demand computing capacity close to their users and data.
- Execution remains critical: Future capacity targets depend on construction, equipment delivery, integration and the ability to operate facilities economically.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
