Pricing
Prices are per GPU hour in USD, for one H100 SXM 80 GB on C1 in us-west. Each GPU comes with 12 CPU cores, 240 GB of system memory and its share of local NVMe. There is no separate charge for CPU, memory, local storage or network inside the rack.
| Plan | USD per GPU hour | Commitment | Notes |
|---|---|---|---|
| On demand | 2.49 | None | Billed per minute, 1 minute minimum |
| Reserved | 1.89 | 1 month or more | Hours held for your team |
| Batch | 1.19 | None, preemptible | Runs when reserved capacity is idle |
Compared to list prices at the four largest providers, updated monthly. The comparison is recomputed with each capacity report.
How billing works
On demand and batch are metered per minute from container start to container exit. Pull time for your image is not billed. Minutes are rounded up per job. A job that runs 61 seconds is billed 2 minutes.
Reserved capacity is billed for the full term at signature, prorated to the day for terms that are not whole months. A reservation of 8 GPUs for 30 days is 8 × 24 × 30 × 1.89, which is 10,886.40 USD. Utilization does not change the invoice. Idle reserved GPUs can be released to batch and earn a credit of 0.60 USD per GPU hour they run batch work.
Shared storage is included up to 5 TB per team. Beyond that it is 0.08 USD per GB per month. Egress is 0.02 USD per GB after the first 1 TB per month. There is no charge for ingress.
Invoices
One invoice per month per team, in USD, payable by card or bank transfer within 30 days. Reserved terms are invoiced at the start of the term. Usage lines are broken down by plan, cluster and day, and match the Usage page in the console to the minute.
Price changes
Prices change at most once a month. A change is announced in the capacity report of the month before it takes effect. Reserved capacity keeps the price it was signed at for the length of its term.
Worked example
A team fine tunes for 3 days on 4 GPUs on demand, then serves an endpoint on 1 reserved GPU for a month, and runs nightly evaluation as batch for 2 hours on 2 GPUs.
Fine tuning: 4 × 72 × 2.49 is 717.12 USD. Serving: 1 × 720 × 1.89 is 1,360.80 USD. Evaluation: 2 × 60 × 1.19 is 142.80 USD. Total for the month, 2,220.72 USD, before storage.
Questions about volume above 8 GPUs go to reserved capacity.