Summary
- Sferical One consists of two NVIDIA DGX SuperPODs, with 16 GB300 NVL72 systems containing 1,152 GPUs in total.
- AstraZeneca, Ericsson, Saab, SEB and Wallenberg Investments founded the venture, while SKF participates as a strategic partner.
- The infrastructure has operated since late summer 2026; its formal 15 October launch marks the progression towards practical business use, not initial commissioning.
Swedish infrastructure venture Sferical AI has formally launched a shared artificial intelligence supercomputer containing 1,152 NVIDIA GPUs, giving five major Swedish industrial and financial organisations access to a common computing platform.
Sferical One was unveiled on 15 October by the consortium founded by AstraZeneca, Ericsson, Saab, SEB and Wallenberg Investments, with SKF participating as a strategic partner.
The system consists of two NVIDIA DGX SuperPODs using a total of 16 GB300 NVL72 systems. The infrastructure has been running since late summer 2026, meaning the launch represents its formal introduction and progression towards practical applications rather than the first commissioning of the equipment.
The participating companies intend to use the platform for workloads ranging from pharmaceutical research and telecommunications engineering to defence applications, financial services and industrial manufacturing.
By pooling investment and access to computing resources, the venture offers an alternative to each participant independently building and operating a comparable AI infrastructure installation.
The project also gives the consortium greater control over selected computing resources located in Sweden, although that arrangement should not be interpreted as eliminating every dependency on international technology suppliers.
Shared infrastructure contains 1,152 GPUs
Sferical AI’s published technical specifications identify a 16-rack configuration built around NVIDIA Grace Blackwell GB300 technology.
Each GB300 NVL72 system contains 72 GPUs, giving the overall platform 1,152 GPUs. SEB describes the installation as having approximately 1,200 GPUs, a rounded figure consistent with the more precise equipment count.
The company’s specification identifies two DGX SuperPODs within the wider installation. That describes the computing architecture rather than 32 separate racks; the 16-rack figure refers to the total published hardware configuration.
SuperPOD systems combine accelerator hardware with the networking and management infrastructure required to support demanding computational workloads. Interconnected processors can work together on tasks too large or complex to be handled efficiently by smaller independent installations.
Such systems also require substantial supporting infrastructure. Electrical distribution, cooling, networking and equipment monitoring must accommodate computing hardware with high power density and demanding availability requirements.
Sferical has not disclosed the facility’s total connected electrical capacity, commissioned IT load, power usage effectiveness or annual energy consumption in its launch statement.
The GPU count therefore cannot be converted into a verified operating megawatt figure. That calculation would require information about the installed equipment, operating conditions and supporting facility systems, rather than assumptions derived from the accelerator specifications alone.
Nor has the company identified the precise facility location within Sweden in the primary announcements reviewed. The infrastructure’s national location is confirmed, but a specific city, data centre building or grid connection should not be inferred.
Consortium separates workloads and data
Sferical AI was established in August 2025 by its five founding organisations. SKF participates through a strategic partnership rather than being identified by the company as one of the original founders.
The ownership and operating arrangements are designed to support shared infrastructure while maintaining separation between participating companies’ data and computing allocations.
SEB confirmed on 15 October that it had begun introducing the infrastructure for selected business applications. The bank described the transition as a move from platform development towards practical use.
The initial applications include tools intended to support information retrieval, data analysis, software development and the use of AI alongside the bank’s internal information.
SEB says each participant can keep its computing allocation and data separate while sharing knowledge and collaborating where appropriate. The statement establishes the intended governance approach, but does not provide an independent technical audit of the platform’s isolation controls.
Those controls are especially relevant where companies operate under different regulatory obligations or handle confidential research, financial information and commercially sensitive industrial data.
SEB has also stated that Sferical is one part of its wider technology arrangements. The bank intends to continue using other platforms and suppliers, selecting services according to business requirements, security, performance and regulatory considerations.
Consequently, the project should not be described as a wholesale replacement for public cloud infrastructure or an arrangement through which every participating business will move all AI workloads onto the Swedish platform.
The consortium has not disclosed the precise capital contributions, ownership percentages or individual computing allocations of its members. Without those figures, the financial benefit of shared infrastructure cannot be independently compared with separate company-owned deployments.
Business applications remain at different stages
The organisations intend to apply the shared platform to different industrial requirements. AstraZeneca is exploring pharmaceutical research and drug discovery, while Ericsson has identified foundation model development for telecommunications.
Saab is examining defence applications, SEB is progressing financial services use cases, and SKF has identified areas including predictive maintenance, quality control and research.
These are application areas described by the participating companies, not independently verified improvements in productivity, research output or commercial performance.
Sferical is also developing services using NVIDIA’s Nemotron open models. The company says the availability of model weights will allow participants to inspect and adapt models for their own operating requirements.
The venture operates Sferical Labs to support technical development and collaboration across the participating businesses. NVIDIA is providing technical expertise and is involved in establishing a technology centre in Sweden.
That collaboration supplements the installed computing platform but does not establish the present scale or operational status of every planned research and support activity.
Sferical has not announced a general commercial colocation offering or published prices for access by unrelated customers. Its confirmed operating model is a shared computing platform developed principally for the participating organisations.
The formal launch establishes a significant infrastructure milestone following the venture’s creation in 2025. Further disclosures will be needed to determine sustained utilisation, facility energy requirements, external customer availability and the volume of production workloads transferred to the system.
