8x H100 and 8x B200 servers.
A whole eight-GPU server, one tenant per box, for teams with a training or serving workload large enough to justify committed capacity. We do not buy this class of hardware speculatively: it is quoted per configuration and provisioned against a signed order.
No capacity available right now. No H100 or B200 server is currently installed in our fleet. Tell us your configuration and term and we will send a quote.
What's on offer
8x H100
- GPUs
- 8x NVIDIA H100
- Tenancy
- Whole server, one customer
- Billing
- Monthly, against a signed order
8x B200
- GPUs
- 8x NVIDIA B200
- Tenancy
- Whole server, one customer
- Billing
- Monthly, against a signed order
Interconnect, storage and network configuration are part of the quote and depend on your workload. We are not publishing generic numbers for hardware that is not in our fleet.
Who this is for
This tier is not a demand test like the RTX 5090 and L40S single-GPU tiers. We buy an eight-GPU server only against a signed commitment, since the purchase is large enough that speculative capacity does not make sense.
Pretraining or large fine-tunes
Multi-GPU jobs that need NVLink-class interconnect between all eight GPUs, not eight independent single-GPU machines.
High-throughput production serving
Sustained request volume that a single-GPU tier can't absorb, where dedicated whole-server capacity is worth committing to.
A fixed monthly budget
Teams that would rather commit to a known monthly cost than meter an hourly rate against unpredictable usage.
Tell us your configuration.
Send your GPU count, term and expected workload, and we will come back with a quote.
Request a quote