64GPUs
- Plan
- Pod
- Lead time
- 3 days
- On-demand
- $2.24
- 3-year rate
- $1.62
Eight nodes on a shared spine. Fine-tuning and research.
Nordlys69.6492° N — Arctic compute
Press M to toggle
Rate card · Valid to 31 March 2027
Two GPUs, three plans, three terms. Eighteen numbers, all-in, fixed for the length of your term.
01Rate cardPer GPU-hour, all-in
Pick a row for the plan, a column for the term. The rate in the cell is the rate on the invoice.
GPU · PlanOn-demandList1 year−18%3 years−28%
H200 · 141 GBLive now · Hall A, Narvik
Pod64 – 512 GPUs
$2.24List rate
$1.84Save 18%
$1.62Save 28%
Dedicated512 – 4,096 GPUs
$2.08List rate
$1.72Save 17%
$1.54Save 26%
Sovereign4,096+ GPUs
$1.98List rate
$1.64Save 17%
$1.46Floor rate
B300 · 288 GBBookable now · racked from Q1 2027
Pod64 – 512 GPUs
$3.09List rate
$2.54Save 18%
$2.24Save 28%
Dedicated512 – 4,096 GPUs
$2.87List rate
$2.37Save 17%
$2.13Save 26%
Sovereign4,096+ GPUs
$2.73List rate
$2.26Save 17%
$2.01Save 26%
Lowest rate on the cardUSD per GPU-hour · EUR at the ECB rate on the invoice date
02Volume stepsBigger clusters, lower rates
The rate drops at four cluster sizes. Lead time grows with the build. Three-year H200 rates shown.
512GPUs
- Plan
- Dedicated
- Lead time
- 11 days
- On-demand
- $2.08
- 3-year rate
- $1.54
Your own fabric. Racked and burned in before handover.
4,096GPUs
- Plan
- Sovereign
- Lead time
- 30 days
- On-demand
- $1.98
- 3-year rate
- $1.46
A private cage, your keys, an on-site team in Narvik.
16,384GPUs
- Plan
- Full hall
- Lead time
- Q3 2027
- Site
- N-02 Kiruna
- 3-year rate
- $1.46
A hall of your own. Floor rate, plus a fixed power price to 2030.
Cluster size Rate per GPU-hour falls with each step
03IncludedWhat's in the rate
No power surcharge, no cooling line, no egress meter inside the EEA. One number per GPU-hour covers all of it.
- 98.7% hydro
Power
Hydro and wind from the NO4 grid, matched hourly.
- PUE 1.08
Cooling
Outside air below 10 °C for 300 days a year. No chillers.
- 3.2 Tb/s
Fabric
Non-blocking InfiniBand inside every pod.
- From 200 TB
Checkpoint storage
An NVMe tier next to the racks, sized to the plan.
- ISO 27001
Security
Biometric halls, ISO 27001, data kept inside the EEA.
- 1 s resolution
Telemetry
Power, temperature and ECC errors for every GPU, live.
- 24 / 7
Hands on site
Engineers in Hall A around the clock, swaps within four hours.
- €0 / GB
Egress in the EEA
Move checkpoints and datasets between EEA regions without a meter.
04VoiceOn predictable cost
Finance teams care less about the lowest rate than about a rate that holds.
Orbit LabsCustomer since 2025
“We budget compute the way we budget rent. One rate per GPU-hour, fixed for three years, and no surcharge when the winter power price spikes.”
—Erik Halvorsen, CFO14invoices, zero variance
Sovereign · 3 yearsRate fixed to 202805BillingReading the card
Need a rate that isn't on the card? Finance answers within one working day.
Sovereign starts at 4,096 GPUs. A whole cage on one fabric is cheaper for us to build and run per GPU, so the rate is lower even with more isolation.
B300 pods are bookable now and racked from Q1 2027. The rate is locked on the day you sign, not the day the racks go live.
The lower rate applies to the whole cluster from the next billing month. Your term end date stays the same.
Yes. Each GPU type is billed at its own row on the card, on one invoice, under one term.
06ContactQ1 2027 allocation is open