Priced by the GPU, not by the guess.
One rate per GPU per hour, no egress fees, no bundling markup. Commit to a term for a steep discount, or stay on-demand and pay for exactly what you use.
| GPU | VRAM | vCPUs | RAM | $/GPU/HR |
|---|---|---|---|---|
| H200 SXM | 141 GB | 208 | 2990 GiB | $4.49 |
| H100 SXM | 80 GB | 208 | 1800 GiB | $3.29 |
| A100 SXM | 80 GB | 240 | 1800 GiB | $2.19 |
| A100 SXM | 40 GB | 124 | 1800 GiB | $1.59 |
| L40S PCIe | 48 GB | 64 | 480 GiB | $0.89 |
Commit for a term, save on every GPU
On-Demand
0%off on-demand
Full flexibility, billed by the minute with no commitment.
- Cancel anytime
- No minimum spend
- Standard support
1-Year Reserved
32%off on-demand
Lock in capacity and a lower rate for a steady workload.
- Guaranteed capacity
- Priority hardware support
- Volume add-ons available
3-Year Reserved
51%off on-demand
Lowest per-GPU rate, for workloads you know aren't going anywhere.
- Deepest discount available
- Dedicated account contact
- Custom cluster sizing
Before you commit to a term
Is pricing really per-GPU, not per-node?+
Yes. A 1x instance is billed at exactly the per-GPU rate shown, and an 8x instance is billed at eight times that rate — no bundling markup for larger configurations.
What's the difference between on-demand and reserved?+
On-demand instances can be launched and terminated anytime with no commitment. Reserved capacity guarantees hardware availability and a locked-in discount for a 1 or 3-year term.
Do you charge separately for storage or networking?+
Local NVMe storage is included with every instance. Network egress has no separate fee. Persistent block storage and object storage are billed separately if you use them.
Can we get a custom quote for a large cluster?+
Yes — for multi-node clusters or dedicated private infrastructure, reach out and we'll put together pricing based on your exact GPU count and term.
Not sure which tier fits your run?
Tell us roughly how many GPU-hours you're looking at and we'll recommend on-demand or a reserved term.
Talk to us