Reserved
Dedicated H200 and B300 pods from 256 GPUs, on one to three-year terms.
GPU halls above the Arctic Circle. Cooled by outside air, powered by rivers, metered to the watt.
Power usage effectiveness, metered every hour across Hall A.
Hydro and wind from the NO4 grid, matched by the hour.
Free-air cooling. No evaporative towers, no chilled water.
Delivered to Narvik’s district heating network last winter.
Two dark-fibre routes south, diverse paths.
Dedicated H200 and B300 pods from 256 GPUs, on one to three-year terms.
Spin up 8 to 512 GPUs in minutes. Billed per second, no commitment.
Private halls under EEA jurisdiction for governments and research labs.

“We trained a 70B model in 19 days and spent a third less on energy. The cold is not a gimmick, it is a cost line.”

“We moved our protein-folding fleet north without changing a line of code. Latency went up two milliseconds. Nobody noticed.”

“A petabyte of satellite imagery a month, and zero litres of water. That number changed how our board talks about compute.”