Cloud GPU computing
A cloud GPU is remote accelerator capacity rented from a provider. You run software on an instance or managed service without buying the hardware. Whether that is useful depends on memory, software, data location, required runtime and the complete bill.
Instances and managed services are different products
An instance exposes a particular CPU, RAM, storage and GPU configuration. Serverless inference bills through a different model, which may include requests, tokens, active time or idle workers. Do not compare a serverless unit price directly with a VM hourly price.
NVGPU's catalog compares 1107 listed NVIDIA configurations across 9 providers and 29 model keys. It does not compare the whole market or promise available stock.
What workloads can use a cloud GPU?
Examples include compatible AI training and inference software, rendering and parallel scientific computing. Suitability depends on the workload and provider environment. Check GPU memory, supported drivers, graphics or compute requirements and any software licensing before renting.
How are cloud GPUs priced?
Start with the full instance rate and billed time. Spot capacity is interruptible; commitments have contractual conditions. Storage and transfer may be billed separately. A listing does not establish how long deployment takes, uptime or job throughput.
Compare catalog-backed prices · Rental checklist · Compare cloud with local ownership
Reviewed September 26, 2026. Sources and methodology · Commercial disclosure.