N-01 · Narvik · Accepting reservations

Reserve a cluster in the cold.

Pick a size, pick a term, and run on GPUs cooled by Arctic air and powered by hydro. Most clusters are live within eleven days.

See the platform
How it works
nordlys-cli — n-01
$nordlys reserve --gpus 1024 --site n-01 --term 1y
✓capacityHall A · 1024 × H200
✓fabric3.2 Tb/s InfiniBand
✓power98.7% hydro · PUE 1.08
→quote$1.38M / month
→first job16 Oct 2026
$
Standard on every cluster
Air-cooledHydro-poweredEEA sovereign
TensorfieldquorumVektorHelix AIKernelsynapseOrbit LabsPhotonTensorfieldquorumVektorHelix AIKernelsynapseOrbit LabsPhoton
01PlatformBuilt into every hall

Six things you get with every Nordlys cluster, whether you rent eight GPUs or a whole hall.

Air-cooled

Outside air below 10 °C for 300 days a year. No chillers to power.

Chillers0

Hydro-powered

Every kilowatt-hour from hydro and wind on the NO4 grid, matched hourly.

Renewable98.7%

Zero water

Free-air cooling means no evaporative towers, in summer or winter.

Water / year0 L

Closed-loop heat

Waste heat leaves at 35 °C and warms Narvik’s district network.

Reused41 GWh

Dedicated fabric

Every pod gets its own 3.2 Tb/s InfiniBand fabric. No noisy neighbours.

HardwareH200 · B300

Live telemetry

PUE, load and carbon metered every hour and exposed through the API.

Uptime99.995%
From$1.62/ GPU-hour, all-in
Request a quote
Outside-air intake louvres on a Nordlys data hallWater-free data hall beside a snowy fjordHeat recovery pipework at the Nordlys campus in Narvik
How it works01/ 03
02How it worksThe cold does three jobs
01 · Air

Air

Outside air below 10 °C for 300 days a year. Fans move it through the halls; there are no chillers to power.

Chillers0
PUE1.09
02 · Water

Water

Free-air cooling means no evaporative towers. The halls use no water for cooling, in summer or winter.

Water / year0 L
Towers0
03 · Heat

Heat

Waste heat leaves the halls at 35 °C and flows into Narvik’s district heating network.

Heat reused41 GWh
Homes4,200
03VoicesFrom the teams on the floor

Two of the labs training in Hall A, in their own words.

Customer since 2024
“Our training runs used to stall on thermal limits every summer. In Narvik they just keep going. Fewer restarts, steadier throughput, and a power bill we can plan around.”
—Amara Okafor, Head of Infrastructure−38%job restarts
Portrait of Amara Okafor01 / 02Live telemetry
Customer since 2025
“We moved 512 GPUs north in eleven days. Same fabric, same tooling, and every kilowatt-hour now shows up as hydro on our carbon report.”
—Jonas Weber, CTO98.7%renewable
Portrait of Jonas Weber02 / 02Hydro-powered
04
FAQ
Before you reserve

Still unsure? An engineer answers within one working day.

Ask a question
How fast can a cluster be ready?

Up to 512 GPUs are ready in about three days. Dedicated pods up to 4,096 GPUs are racked and burned in within eleven days. Larger builds are scheduled with you.

In Norway, inside the EEA. Data never leaves the hall unless you move it, and sovereign tenants hold their own encryption keys.

NVIDIA H200 today, with B300 pods bookable from Q1 2027. Every pod sits on a dedicated 3.2 Tb/s InfiniBand fabric.

Per GPU-hour, all-in: power, cooling, fabric and checkpoint storage are included. Longer terms lower the hourly rate.

Yes. Run Slurm, Kubernetes or bare metal, with your own images. Our engineers help with the first deployment.

05ContactQ1 2027 allocation is open

Your next run starts in Narvik.

Tell us cluster size and start date. An engineer replies within one working day.

Plan capacity