Published 2026-10-16 · Event 2026-10-15 · The editorial team

Sferical One is running; its data-centre load remains undisclosed

The shared Swedish AI supercomputer began running in late summer and was launched publicly on 15 October. Its partners have started use cases, but the sources do not state the facility's IT load, electrical demand or data-centre site details.

Sferical AI said on 15 October 2026 that Sferical One, a shared AI supercomputer for AstraZeneca, Ericsson, Saab, SEB and Wallenberg Investments, had been launched. SKF is a strategic partner. The company says the system had already been running since late summer, so the launch date is not its commissioning date.

The release describes an NVIDIA DGX SuperPOD built on 16 NVIDIA DGX GB300 systems. Sferical's stated areas of early work include drug discovery, telecommunications foundation models, defence applications, financial services and industrial uses such as predictive maintenance and quality control. These are partner use areas, not published performance results or customer deployment volumes.

SEB separately said on 15 October that it had started using the Swedish infrastructure for selected business applications. Its notice describes an initial focus on ready-to-use tools for staff, information search, data analysis and software development. It also says SEB will continue using multiple AI platforms, selecting among them according to business value, security, performance and regulatory needs. The shared system is one part of a wider sourcing strategy, not a replacement for every provider.

For data-centre developers, the launch is evidence of an operating Swedish AI-compute platform and a shared-infrastructure model backed by several industrial companies. It does not disclose a building address, operator-level facility specifications, IT load, utility connection, electricity consumption, cooling design or heat-reuse arrangement. GPU counts and server-system descriptions cannot be converted into site MW without the workload, rack configuration, redundancy and facility boundary.

A project team assessing similar high-density workloads should request the exact boundary for each number: IT equipment load, total facility demand, contracted connection capacity and expected annual energy use. It should also ask for rack-level power and cooling requirements, backup design, commissioning evidence and the responsible data-centre operator. The existing guides on high-density racks and liquid cooling outline the technical questions to carry into site screening.

Sferical AI was founded in 2025 by AstraZeneca, Ericsson, Saab, SEB and Wallenberg Investments. Its original announcement said the planned system was intended to be located in Linköping. The October launch release says only that Sferical One is in Sweden; the material reviewed here does not confirm the current facility address, so this article does not place it on the site's project map or attribute it to a named campus.

Why it matters

Sferical One shows that Swedish industrial demand is already using shared AI compute, while leaving the facility's power and cooling requirements open for site-level due diligence.

What remains unknown

The public material reviewed does not identify the current facility address, site operator, IT load, facility demand, grid connection terms, annual energy use or cooling and heat-reuse systems.

Sources