Rent dedicated GPUs for your AI workloads
Deploy high-performance AI infrastructure in seconds. Access premium NVIDIA hardware without long-term commitments or hidden fees.
Available Hardware
PUBLISHED HOURLY RATES · CONFIRMED ON REQUEST
Loading live GPU catalog
PLATFORM ADVANTAGES
- No long-term lock-in
- Scale on demand
- Per-second billing
- Dedicated InfiniBand
Running in four steps
No procurement cycle, no infrastructure team required.
Pick your hardware
Choose H100, H200 or B200 and the allocation window that matches your run.
Reserve the cluster
Capacity is held while you confirm. Nothing bills until the instance starts.
Connect and run
Standard tooling, dedicated InfiniBand between nodes, no proprietary runtime to adopt.
Scale or stop
Add nodes mid-project or release everything — billing stops at the second.
Hardware reference
Not sure which generation you need?
Ampere through Blackwell, memory per accelerator, and how NVLink and MIG decide whether a cluster can be partitioned or scaled — the full comparison sits on the Buy GPU page and applies to rented capacity just the same.
Before you reserve
What is the difference between H100 and H200?
Both deliver 3,958 TFLOPS at FP8. The H200 has 141GB of memory and 4.8 TB/s of bandwidth versus 80GB and 3.35 TB/s, so it suits models that are memory-bound rather than compute-bound.
Is B200 available now?
B200 is early access. Pricing is by enquiry and allocation is limited — talk to sales about your timeline.
How does per-second billing work?
You are charged for the exact runtime of the instance. There is no rounding up to the hour and no minimum charge.
Can I move to owned hardware later?
Yes. Rented usage can be credited against a purchase — see the Buy GPU page for the trade-offs between OpEx and CapEx.
Reserve capacity for your next run
Per-second billing, no minimum term, cancel whenever the job finishes.