If you’ve ever had a training run die halfway through because capacity disappeared, you already get the pain: long-running AI workloads don’t mix well with “best effort” GPU availability.

That’s why we’re rolling out GPU Reservations on Hyperbolic — a new way to reserve and guarantee GPU availability for up to 30 days, designed specifically for multi-day or multi-week training, fine-tuning, and inference workloads.

No multi-month contracts. No uncertainty. Just locked-in capacity when you need it.

Why GPU Reservations

Most teams running serious workloads face the same tradeoff:

  • On-demand is flexible, but not always reliable for long jobs.

  • Long-term commitments guarantee capacity, but lock you in for months.

GPU Reservations gives you the best of both: guaranteed GPUs + discounted pricing, with 7 / 14 / 30-day options and no multi-month commitment.

What you get

When you reserve GPUs, you get:

  • Guaranteed capacity for the full duration of your rental

  • A dedicated Slack channel with our engineering team

  • Dedicated support for your reserved instance

  • Custom storage volumes for your workload

How it works

Making a GPU reservation is straightforward:

  1. Choose “Reserve” when launching a GPU

  2. Configure your setup: GPU type, region, and count (up to 32 GPUs)

  3. Pick a duration: 7 days, 14 days, or 30 days

  4. Prepay and launch instantly

That’s it — your capacity is locked.