On-demand workspace
Run your own image and commands. Connect with SSH, JupyterLab or a browser terminal. Release the GPU when the job is done.
GPU runtime + persistent storage
Explore GPU workspaces →Reserved GPU nodes
Whole GPU nodes. Reserved for your team.
Keep dedicated capacity for recurring workloads. Run GPU workspaces, model instances, clusters and fine-tuning on nodes assigned exclusively to your account, for a term agreed with our team.
Whole nodes · Agreed region and term · Contract pricing
Your capacity
One reservation. The tools you already use.
For teams that need dedicated nodes for a continuing workload, with capacity and costs arranged in advance.
Nodes assigned to your account
Other customers do not run workloads on your reserved nodes. Launch up to the free capacity in your reservation; there is no automatic overflow into the public pool.
Workspaces, serving and training
Use the same console, Compute APIs and templates. Your GPU instances, clusters, model instances and fine-tuning jobs run on your reserved nodes.
Storage included by contract
Agreed cloud-drive and shared-filesystem quotas in the reserved region are covered by the period charge. Request an amendment when you need more storage.
From plan to workload
Agree the capacity. Activate it. Launch as usual.
- 01
Plan with our team
Share your GPU type, preferred region, whole-node count, expected term and storage requirements. We agree capacity, pricing and payment terms with you.
- 02
Start on the agreed terms
For first-invoice contracts, the reservation normally starts when that invoice is paid. Other funding and activation arrangements are defined in your contract.
- 03
See and use your nodes
The console's Reserved Nodes page shows your capacity, workloads and storage quota. Account owners can also review the payment schedule.
Reservation charges are for contracted capacity, including idle capacity, rather than each workload's runtime. Platform model API calls remain separately billed.
Choose your setup
A workspace, an endpoint, or whole nodes?
Dedicated inference
Deploy a model as an endpoint on dedicated GPUs. Call it through the unified API instead of managing a workspace yourself.
GPU runtime + persistent storage
Explore model endpoints →Reserved GPU nodes
Keep whole nodes exclusively for your account over a contracted term. Run your workspaces, model instances, clusters and training there.
Contracted capacity + storage quotas
Explore reserved nodes →Before you reserve
Capacity, payments and storage
- How do I reserve GPU nodes?
- Contact sales with your GPU type, preferred region, node count, term and storage needs. Reservations are arranged by contract, not purchased instantly in the console. Pricing depends on the agreed capacity, term and payment schedule.
- Is this the same as a dedicated inference endpoint?
- No. A dedicated inference endpoint serves a model. A GPU reservation assigns whole nodes to your account for a contracted term. You can run GPU workspaces, clusters, model instances and fine-tuning jobs on those nodes, using the existing console, templates and APIs.
- What happens when all my reserved GPUs are busy?
- New workloads wait for you to free capacity or arrange more nodes. A launch is refused when your reservation has no free GPUs; it does not automatically spill into the public on-demand pool. Other customers cannot use your reserved nodes.
- When does the reservation start?
- For contracts using a first invoice, the term normally starts when that invoice is paid. The first invoice includes the advance, if any, plus the first period. Other payment arrangements, including use before first payment, must be agreed in the contract.
- Are storage and model API calls included?
- Cloud-drive and shared-filesystem quotas in the reserved region are included as agreed in your contract. Quota increases require an amendment. Platform model API calls remain separately billed; a node reservation does not include unlimited tokens or media generation.
- What happens at the end of the term?
- Workloads on the reserved nodes stop when the term ends. Cloud drives and shared filesystems are kept and then billed at standard storage rates. Renew the reservation or save your results and delete storage you no longer need.
Plan your reserved GPU capacity
Tell us the workload, region, node count, term and storage you have in mind.