Press M to toggle
After five months on a dedicated 2,048-GPU pod, Tensorfield is moving all of its frontier training to N-01 and has reserved B300 capacity for 2027.
Tensorfield, which trains open-weight foundation models for European enterprises, will run all of its frontier training in Hall A in Narvik. The company has extended its Dedicated pod contract to three years and reserved B300 capacity from Q1 2027.
Tensorfield's 2,048 H200s went live on 3 March 2026. Since then the team has trained two 70B models on the pod. Energy cost per run is down 34% compared with its previous region, and job restarts are down 38%.
Our training runs used to stall on thermal limits every summer. In Narvik they just keep going.
The remaining training workloads move north over the autumn. Inference stays close to Tensorfield's customers in central Europe. The full story, with numbers from the first six months, is in the Tensorfield case study.
Run on Arctic air and hydro from $1.46 per GPU-hour. An engineer replies within one working day.
