Dedicated GPU capacity,
without competing for it.
A dedicated cluster, sized and placed where you choose, so you never wait for allocation on a shared cloud. You get the margin of a long-term rate, and rollover keeps you on the newest generation.
Long-term margin without falling a generation behind.
A long-term rate usually means staying on the GPU you started with. Here, part of the commitment can move to the next generation when it ships, so the rate and the hardware both stay competitive.
- Capacity you don't have to findThe cluster is reserved for you for the whole term. No waiting for allocation, and no other tenant on your nodes.
- Your size and regionStart at 8 nodes or one NVL72 rack in the region you choose, and add capacity in monthly waves of 32 nodes as traffic grows.
- Margin on every GPU-hourA long-term rate is lower than on-demand pricing, and it is fixed for the term. The difference stays with you on each hour you serve.
- Current hardwarePart of the commitment can move into a newer GPU generation when it ships.
From a sizing call to serving on your own cluster.
Step 1: Size it to the workload
Tell us the models, the training or inference load and the region. We recommend a GPU and a starting size.
You getSizing and a GPU recommendationStep 2: Sign and prepay into escrow
Proposal, then MSA. Delivery dates run from signing.
You getA dated startStep 3: Take the first wave
Your first nodes or rack are burned in and handed over, as bare metal or managed.
You getA cluster to run onStep 4: Ramp up in waves
Further capacity arrives in monthly waves of 32 nodes. Later in the term you can also roll over or exit.
You getA dated ramp
The terms AI teams ask about.
Every cluster follows the same contract. Rates are on the pricing page.
- Start
- 8 nodes or one NVL72 rack
- Growth
- In monthly waves of 32 nodesDates counted from signing.
- Modes
- Bare metal or managedSame single-tenant hardware either way.
- Exit
- Exit option in the contractYour proposal sets the terms.
- Rollover
- Into a newer GPU generationPart of the commitment can move into a newer GPU generation when it ships.
Questions buyers ask first.
Something missing? Ask us directly. We usually reply within a few hours.
On-demand GPU capacity is priced for flexibility, and you pay for it on every hour. A reservation trades that flexibility for a lower rate, fixed for five years. The rate for each GPU is on the pricing page, so you can compare it with what you pay today.
The contract includes an exit option, on terms set out in your proposal. We can also help place capacity you aren't using with other buyers, though we don't guarantee a buyer.
Both run on the same single-tenant hardware. Bare metal gives you root on every node with your own OS image and scheduler. Managed means we also run the software layer on top, such as a scheduler for containers and jobs, or inference endpoints.
8 HGX nodes (64 GPUs) or one NVL72 rack (72 GPUs). NVL72 is reserved only as whole racks.
Part of the commitment can move into a newer GPU generation when it ships.
The contract includes an exit option, on terms set out in your proposal.
HGX B300 takes about 10 to 14 weeks from signing. Other GPUs are dated in the proposal, and larger clusters arrive in waves.