What you pay for, before what it costs
GPUs are billed by the hour and storage by the GiB, both published. VM Clusters and Provisioned Throughput start from their billing unit; the terms are agreed with our team.
- Per-second billing
- Reserved terms available
- SOC 2 Type II
What gets billed
Each product is billed in its own unit. Start with the unit, then go where you need.
Container Compute
per GPU, per hour
Quoted with sales
Metered by the second, with no quota to request first.
See pricingVM Cluster
per node, on a term
from $1.48
A dedicated multi-node cluster. Terms follow the volume and the length.
See pricingProvisioned Throughput
per PTU, per minute
$0.05
The same rate on every model. PTU count and term are agreed with sales.
See pricingVM Cluster
Rent whole nodes as VMs with root access on every one. There is no self-serve rate: the price follows the GPU model, the node count, and the term.
| GPU | VRAM | Architecture | Per node, on a term |
|---|---|---|---|
| NVIDIA B300 | 288GB | Blackwell | Quoted by term |
| NVIDIA B200 | 192GB | Blackwell | Quoted by term |
| NVIDIA GB300 | 288GB | Blackwell | Quoted by term |
| NVIDIA H200 SXM | 141GB | Hopper | Quoted by term |
| NVIDIA H100 SXM | 80GB | Hopper | Quoted by term |
- Eight GPUs per node; GB300 comes as an NVL72 rack
- InfiniBand between nodes, a dedicated public IP per node
- Root SSH, a 2 TiB boot disk, optional NFS shared storage
- Prepaid reserved contract; volume and term set the rate
Container Compute
Run Workspaces and Jobs as containers on shared clusters, metered by the second.
| GPU | VRAM | Architecture | On-Demand | |
|---|---|---|---|---|
NVIDIA H100 SXMSelf-serve | 80GB | Hopper | $2.98/hr | Start now |
NVIDIA A100 SXMSelf-serve | 80GB | Ampere | $1.48/hr | Start now |
NVIDIA L40SSelf-serve | 48GB | Ada Lovelace | $1.80/hr | Start now |
Estimate your monthly cost
Live estimates from published hourly rates. Adjust the GPU, count, usage, and storage.
Warm
Cluster Storage: fast storage for active datasets and checkpoints during training.
Cold
Object Storage: low-cost storage for archives and long-term data.
- Compute
- $2,175/mo
- Storage
- $0.00/mo
- Total
- $2,175/mo
Estimates only, at published hourly rates. Final pricing may vary; reserved terms are quoted with sales.
Provisioned Throughput
Contract throughput instead of running the model yourself. Inside your contracted capacity you are billed on PTUs rather than tokens.
$0.05
per PTU, per minute
$2,190
per PTU, per 730-hour month
Arranged with sales
PTU quantity and contract term
Cluster and Object Storage
Persistent storage for datasets, checkpoints, and artifacts, billed daily on what you actually store.
Fast shared working sets for active runs: code, environments, and hot datasets.
Low-cost, durable storage for checkpoints, logs, and long-term archives across clusters.
Billed daily on the data actually stored, not provisioned capacity.
Frequently Asked Questions
What's the difference between On-Demand and Reserved?
On-Demand is pay-as-you-go capacity billed per second: spin up a GPU when you need it and release it when the run ends. A100, H100, and L40S run on-demand. Reserved holds capacity for your timeline, at a rate set by term and volume. A100 and H100 can be reserved; L40S is on-demand only. H200 and the Blackwell GPUs are reserved only.
What's the difference between Cluster Storage and Object Storage?
Cluster Storage allows you to share files across multiple workloads and provides faster network performance, ideal for collaborative training jobs. Object Storage is best for storing large datasets and artifacts at a lower cost.
What's included in a reserved term?
Guaranteed capacity on A100, H100, H200, B200, B300, or GB300 for the length of your term, with dedicated support and volume discounts.
Where AI models
get their GPUs
- Start in minutes
- Scale to multi-node clusters
- Capacity reserved to your timeline
- Dedicated support