GPU cloud for training and inference
On-demand NVIDIA GPU instances through the QuantaCloud console, plus reserved and dedicated capacity configured for production workloads at scale.
01 / On-demand
Live catalogChoose an instance, add a payment method, and deploy from the console. Compute is metered by the second.
02 / Reserved
Planned capacitySecure dedicated nodes, multi-node deployments, and term pricing with a configuration scoped to your requirements.
One path to production
Prove the workload on an on-demand instance, then secure the capacity it needs for production. QuantaCloud gives you a clear path between both buying models, one commercial relationship, and one team to call when the requirements change.
Launch on-demand compute without waiting for a quote or contract. Published pricing makes the cost clear before the instance starts.
Explore on-demand GPUs →When the workload needs guaranteed capacity, longer terms, or a custom topology, move into a dedicated configuration with our team.
Start a capacity plan →Create an account and launch from the console. Save the sales process for when the workload actually needs it.
Published hourly rates and per-second metering make on-demand spend legible before and after every run.
Use what you learn on-demand to plan dedicated capacity, volume pricing, and the right production configuration.
On-demand compute
Current availability, billed by the second. Prices are shown per GPU, per hour.
| GPU | VRAM | Per node | Price |
|---|---|---|---|
| NVIDIA H100 | 80 GB | up to 4× | $2.59/GPU/hr |
| NVIDIA RTX PRO 6000 Blackwell | 96 GB | up to 4× | $2.38/GPU/hr |
| NVIDIA DGX A100 | 80 GB | 1× | $1.50/GPU/hr |
| NVIDIA A100 SXM | 80 GB | up to 8× | $1.49/GPU/hr |
| NVIDIA A100 | 80 GB | up to 4× | $1.48/GPU/hr |
| NVIDIA L40S | 48 GB | up to 8× | $1.09/GPU/hr |
| NVIDIA L40 | 48 GB | up to 4× | $0.94/GPU/hr |
| NVIDIA RTX 6000 Ada | 48 GB | up to 8× | $0.78/GPU/hr |
| NVIDIA RTX A6000 | 48 GB | up to 8× | $0.48/GPU/hr |
Need guaranteed capacity, a custom configuration, or volume terms? Plan a dedicated cluster.
Reserved and dedicated
For production training, sustained inference, and workloads that cannot depend on on-demand availability. Start with requirements—not a generic bundle.
Tell us the GPU type and count, deployment region, interconnect, start date, and expected term. Rough requirements are enough to begin.
We return with availability, a proposed configuration, deployment timing, and commercial terms in writing.
We coordinate provisioning and bring the environment online with agreed access and support.
Optional
Add monitoring, incident response, workload migration, and a named technical contact to the capacity plan. The access boundaries and operating responsibilities are agreed before launch.
Still validating? Start on-demand instead.
Before you launch
Need an answer specific to your architecture? Ask our team.
Capacity planning
Share what you know today. We will reply within one business day with the next concrete step.