Sferical AI launched the Sferical One supercomputer on October 15, 2026, a shared AI system in Sweden built for AstraZeneca, Ericsson, Saab, SEB and Wallenberg Investments, with SKF as strategic partner. According to Sferical AI’s press release, the system is an NVIDIA DGX SuperPOD installation built on DGX GB300 systems, and it has been running since late summer 2026.
Instead of each company building its own cluster, the partners pool the investment and use one machine. Sferical AI says the companies are now starting to apply it to their own work, from drug discovery to telecom foundation models.
On this page
The project as Sferical AI and NVIDIA state it
| Item | As announced |
|---|---|
| Operator | Sferical AI (Sferical AB), Stockholm |
| Founding partners | AstraZeneca, Ericsson, Saab, SEB, Wallenberg Investments |
| Strategic partner | SKF, since April 30, 2026 |
| Location | Linköping, Sweden |
| System | Two NVIDIA DGX SuperPODs, 16 racks of NVIDIA DGX GB300 (GB300 NVL72) |
| GPUs | 1,152 |
| Status | Launched October 15, 2026; running since late summer 2026 |
| Software | Services built on NVIDIA Nemotron open models |
The hardware figures come from Sferical AI’s About page, which lists two DGX SuperPODs in a 16-rack GB300 NVL72 setup with 1,152 GPUs, and from NVIDIA’s October 14 Nordics post, which gives the same configuration. Sixteen racks of 72 GPUs each make 1,152. The site in Linköping is named in Sferical AI’s April 30, 2026 SKF announcement, which refers to “Sferical AI’s infrastructure in Linköping”. Neither company has published a power figure in megawatts, so none is given here. NVIDIA’s published specifications for the rack are on our GB300 NVL72 page.
What the partners plan to run on it
Sferical AI lists the first areas of work: drug discovery at AstraZeneca, foundation models for telecommunications at Ericsson, defence applications at Saab, financial services at SEB, and predictive maintenance, quality control and R&D at SKF. It also plans services based on NVIDIA’s Nemotron open models; because the weights are open, members can inspect and fine-tune them and run them in environments that meet their own security requirements.
“Sferical One enables the companies to train and run AI models at a larger scale on a sovereign platform,” Jenny Nordlöw, CEO of Sferical AI, said in the release.
Alongside the machine, NVIDIA is setting up a Technology Center in Sweden that supports the consortium through its solution architects, according to Sferical AI. The company also runs Sferical Labs, where the partners and Sferical develop AI applications together.
From announcement to launch
Sferical AI was set up on August 22, 2025, when the five founders said the system was intended to be located in Linköping and would use two DGX SuperPODs with DGX GB300 systems and 1,152 GPUs in total. The launched configuration matches that plan. SKF joined as strategic partner on April 30, 2026, and the system went into operation in late summer, before the formal launch in October.
The launch was marked in Stockholm during NVIDIA’s European tour before GTC Berlin. NVIDIA’s GTC Berlin news blog reported on October 15 that Jensen Huang, Marcus Wallenberg, Jenny Nordlöw and Daniel Ek discussed the system on stage at the Eric Ericsonhallen.
At 1,152 GPUs, Sferical One is far smaller than the gigawatt campuses that dominate AI infrastructure news. It follows a different model: one machine shared by a fixed group of industrial companies, rather than a cloud region or a single lab’s training cluster. Other announced projects are tracked in AI data centers, and more project news is in AI data center news.
