Summary

• SWI Group has launched SWI Compute with reported access to 3.6GW of power capacity across 24 sites in Europe and the US.
• The first UK facility is planned to begin at 30MW by late 2027, with a longer-term campus target of 530MW.
• The company reports approximately $1 billion of NVIDIA equipment orders and initial customer contracts, but hardware delivery, commissioning and utilisation remain to be demonstrated.

Amsterdam-listed infrastructure investor SWI Group has launched a specialist AI computing business with plans for an initial 30MW facility in the United Kingdom by the end of 2027, combining its existing data centre development interests with access to NVIDIA computing systems.

The new company, SWI Compute, says it is backed by 3.6GW of power capacity secured by its parent across 24 sites in Europe and the United States. It has also placed orders worth approximately $1 billion for NVIDIA systems and signed initial contracts with customers.

The planned UK facility would form the first phase of a campus that SWI eventually intends to expand to 530MW. The company is targeting 2.3GW of AI factory capacity by 2029, although the figures describe proposed development rather than computing infrastructure currently available for commercial use.

Announced on 26 October, the launch reflects a strategy of connecting land, electrical capacity, data centre construction and computing services within one group. SWI argues that control over those resources will help it deliver AI infrastructure more predictably than businesses attempting to acquire them after customer demand has emerged.

Power access underpins the commercial model

AI computing infrastructure places substantial demands on electricity supplies, cooling equipment and physical space. The availability of advanced processors is commercially useful only when those devices can be installed and operated within facilities that support their electrical and communications requirements.

In parts of Europe, suitable data centre developments face long lead times associated with grid connections, planning decisions and the construction of supporting infrastructure.

SWI says its parent began investing in land, power and data centre development opportunities in 2020, building a portfolio intended to support long-term growth in digital infrastructure.

The company describes the resulting 3.6GW as secured power capacity, distributed across sites in Europe and the United States.

That figure should not be interpreted as electricity currently available to completed AI computing facilities. Securing capacity through development arrangements differs from operating a commissioned site with an active grid connection and installed equipment.

The practical value of a power allocation depends on the relevant contractual conditions, connection timetable, network upgrades and costs associated with preparing the facility for operation.

These distinctions are important when comparing infrastructure developers. A group may control valuable development rights while still requiring substantial capital expenditure before customers can use the computing capacity.

The UK project starts at 30MW

SWI Compute plans its first large AI factory in Britain, with a 30MW opening phase expected by the end of 2027.

The company describes an eventual 530MW campus, representing a substantial increase from the planned initial installation. Achieving that scale would require additional electrical infrastructure, construction work, computing equipment and customer demand.

Industry reporting associates the development with SWI’s planned Cambridge data centre project. However, the company’s launch material does not identify the precise UK site, and the relationship between the reported Cambridge development and the full 530MW ambition has not been independently reconciled.

The location and subsequent expansion phases should therefore be regarded as subject to further confirmation.

SWI has not published a complete construction budget for the UK campus, a detailed schedule covering the proposed expansion or independently verified confirmation that every future phase has the necessary permissions and utility arrangements.

Those details affect the financial and operational feasibility of a development whose eventual proposed electrical demand would be substantial.

Additional considerations include the availability of cooling infrastructure, high-capacity network connections and specialist engineering resources.

Construction timing will also matter because AI computing hardware changes more quickly than the buildings and electrical systems required to accommodate it.

Hardware commitments require careful interpretation

SWI Compute has joined NVIDIA’s cloud partner programme and intends to build its computing facilities around the chipmaker’s reference architectures.

The company describes its planned 30MW UK configuration as incorporating approximately 7,000 Rubin GPUs and 3,500 Vera CPUs within Vera Rubin NVL72 systems, connected through advanced NVIDIA networking technology.

However, the same announcement states that the first hardware deployment will use NVIDIA Blackwell Ultra systems, with Vera CPU racks following subsequently.

Those statements indicate a phased technology plan, but the precise relationship between the initial Blackwell Ultra installation and the proposed Rubin-based configuration is not fully explained.

It would therefore be misleading to suggest that all of the planned Rubin processors have already been delivered or that the first 30MW installation is commercially operating.

The company has also identified NVIDIA Quantum-X800 InfiniBand and Spectrum-X Ethernet technologies for communications between computing systems.

High-performance networking is particularly important when AI workloads involve large numbers of processors operating together. The ability to move information between computing devices can affect overall performance, even where the individual processors are highly capable.

Reference architectures can help standardise combinations of computing, networking, power and cooling equipment, reducing some of the engineering work associated with designing each facility.

Nevertheless, published reference performance does not guarantee the efficiency achieved within a particular data centre. Actual results depend on workloads, cooling arrangements, utilisation and the integration of equipment.

Equipment orders are not operating revenue

SWI Compute says it has ordered approximately $1 billion of NVIDIA systems and finalised contracts with initial customers, including an unnamed US-based AI inference business.

The announcement does not disclose the full terms of the equipment commitments, the financing arrangements or the timetable under which individual systems will be delivered.

It also provides limited information about the customer contracts, including their values, duration and minimum utilisation requirements.

Those details influence the economics of an AI cloud business because advanced processors require substantial upfront investment and can become commercially less competitive as newer generations arrive.

Customers may purchase dedicated computing capacity or use services on demand, with different arrangements creating different risks for the infrastructure operator.

Longer-term commitments can improve revenue visibility, but they may also require guaranteed service capacity and performance. On-demand services offer flexibility to customers while leaving operators exposed when expensive equipment is underutilised.

SWI Compute intends to support workloads ranging from model training and inference to scientific and industrial computing.

Maintaining high utilisation across those activities requires suitable software, networking and operational support. Different workloads may place different demands on processors, memory and data transfer, making broad compatibility commercially useful but not sufficient to establish profitability.

The group also presents European operation as an advantage for customers concerned about digital sovereignty. The location of infrastructure can support particular data handling requirements, but customers still need to examine ownership, contractual access, technical dependencies and the arrangements governing support services.

Expansion requires operational proof

SWI Compute’s strategy combines two capital-intensive activities: developing large infrastructure sites and operating computing hardware whose commercial value can change substantially over a relatively short period.

Vertical integration may reduce dependence on external landlords or infrastructure suppliers, but it also exposes the company to construction costs, electricity market conditions and the financial risks associated with purchasing specialist processors.

Scaling across multiple jurisdictions introduces additional complexity. Grid arrangements, planning rules, labour availability and customer requirements can differ considerably between countries.

The company reports having secured its initial customers and equipment orders, providing a commercial foundation for development. However, those commitments do not establish that the proposed sites have been built, that computing systems are installed or that contracted services have entered production.

The first important milestones will be confirmation of the UK site’s final location, completion of the initial infrastructure, delivery of the planned hardware and commencement of customer workloads.

Further evidence will be needed to evaluate operating efficiency, utilisation, financial returns and the feasibility of expanding the campus towards its much larger stated capacity.

SWI Compute has outlined an ambitious route into European AI infrastructure, supported by an existing development portfolio. The commercial test will be whether its access to power and equipment can be converted into operational computing capacity on the proposed timetable and at sustainable cost.