Reserved GPUs
Guaranteed GPU capacity,
reserved for your team.
Commit to NVIDIA A100, H100, and more on weekly, monthly, or annual terms. Guaranteed availability for long runs, negotiated rates below on-demand, and a team that helps you spec the right hardware, which you book from us.
The problem
For anything longer than a quick experiment, on-demand alone stops being enough.
On-demand capacity is never guaranteed. When you are mid-way through a multi-week training run, a GPU shortage can stall your roadmap for days.
Committed contracts at the hyperscalers are rigid and expensive. Long minimum terms, opaque pricing, and penalties make it hard to right-size.
Cost planning is guesswork. Fluctuating spot prices and surprise egress bills make it almost impossible to forecast a quarter of GPU spend.
Getting the wrong hardware wastes money. Over-provisioning H100s for a job an A100 handles fine, or the reverse, quietly burns budget.
How PodStack solves it
Reserve the capacity you need, at a rate you can plan around, on hardware we help you choose.
Reserve the exact GPUs you need for the length of your commitment. The hardware is yours for the term, so long runs never get pre-empted by a shortage.
Weekly, monthly, and annual terms unlock rates well below on-demand. The longer and larger the commitment, the better the effective price per GPU-hour.
Choose across NVIDIA A10G, L40S, A100 40GB / 80GB, and H100. Our team helps you spec the right hardware for your workload, which you then book from us.
A fixed reserved rate plus per-second billing on anything on-demand, with zero egress fees. Your GPU spend becomes a number you can plan around.
Reserved capacity runs on the same PodVirt stack as the rest of PodStack, in our ISO 27001 certified data centres, with the option to self-host via the licensable platform.
Specs and details
Frequently asked questions
What is a reserved GPU and how is it different from on-demand?+
A reserved GPU is capacity you commit to for a fixed term - a week, a month, a quarter, or a year. In exchange for the commitment you get guaranteed availability for the whole term and a lower effective rate than on-demand. On-demand is billed per second with no commitment, but capacity is not guaranteed for long runs. Many teams reserve a baseline of GPUs and burst on-demand for spikes.
Which GPUs can I reserve?+
You can reserve NVIDIA A10G, L40S, A100 40GB, A100 80GB, and H100 80GB, as fractional slices or as full and multi-GPU nodes. If you are not sure which fits your workload, our team helps you spec the right hardware, which you then book from us.
How is reserved capacity priced?+
Pricing is sales-led and depends on the GPU model, quantity, and term length. Longer and larger commitments unlock better per-GPU-hour rates. There are no egress fees. Talk to our team with your requirements and we will size a quote to your workload.
Can I still burst beyond my reservation?+
Yes. Your reserved GPUs give you a guaranteed baseline, and you can burst onto on-demand capacity billed per second whenever you need more, so you are never capped by your commitment.
Talk to our team
Tell us your GPU model, quantity, and how long you need it. We will spec the right hardware and send back a reserved-capacity quote sized to your workload.