The Nordics and NVIDIA Show What’s Possible When a Region Commits to Advanced Computing

NVIDIA highlighted Nordic AI projects ahead of its GTC Berlin keynote. In Sweden, a consortium anchored by Ericsson, AstraZeneca, Saab, SEB and Wallenberg Investments launched Sferical One, a shared supercomputer with 1,152 GPUs across two DGX SuperPODs. NVIDIA says AI cloud capacity across the region is projected to reach 1 gigawatt by the end of 2027.

Company examples include Spotify using TensorRT-LLM for podcast recommendations, ABB Robotics' simulation tool, Elisa's meeting assistant, Swedish speech lab Klang's open Pianissimo transcription model, and Einride building autonomous freight trucks on NVIDIA Hyperion.

Strip out the marketing gloss and there's a real pattern: Nordic industrial heavyweights pooling money for shared compute, and small startups post-training open models for narrow, valuable jobs like patent analysis and Swedish transcription. That's a healthy way for a mid-sized region to compete.

The open Swedish transcription model on Hugging Face is the kind of public good we like to see come out of these programs.

The caveat is that this is a vendor blog, and the performance claims — 99% simulation accuracy, 80% faster commissioning — are company figures. And as with Europe broadly, regional sovereignty here is being built on one supplier's hardware.

I'm watching what Jensen Huang announces at GTC Berlin, and whether that gigawatt of capacity arrives on schedule given grid constraints.