GPU Cloud

Launch fast, pay less

A GPU in 60 seconds. 20,000+ cards across the marketplace, ready from the dashboard or fully automated through the API, CLI and SDK.

Get started See live pricing

Pricing

No guesswork. Just GPUs.

Start with $5. Prices move with supply and demand across the whole marketplace, and every offer is queryable before you commit — no quotas, no contracts, no sales call to find out the number.

GPU pricing Talk to us

RTX 4090 Ada Lovelace · 24 GB $0.35/hr RTX 5090 Blackwell · 32 GB $0.41/hr RTX PRO 6000 WS Blackwell · 96 GB $1.07/hr RTX PRO 6000 S Blackwell · 96 GB $1.33/hr H200 Hopper · 141 GB $3.53/hr B200 Blackwell · 192 GB $4.95/hr See all 52 models →

The fleet

Your infrastructure launchpad

Train, fine-tune and serve across one of the larger NVIDIA fleets on the market. Take the newest silicon when you need throughput, take last generation when you need volume — and hand it all back the moment the job ends.

Start building API gateway

Blackwell

B200 192GB RTX PRO 6000 S 96GB RTX PRO 6000 WS 96GB RTX 5090 32GB

Hopper

H200 141GB H100 SXM 80GB

Ada Lovelace

RTX 4090 24GB L40S 48GB

Ampere

A100 40 / 80GB RTX 3090 24GB

What you get

Built for people who ship

On-demand deployment

Spin up 4090s, H200s, B200s and more on your timeline. No upfront negotiation, no quota request, no waiting list.

Transparent pricing

Per-second billing on on-demand, interruptible and reserved capacity. A $5 minimum is the whole commitment.

Isolated instances

Dedicated hardware with full environment control. No container sharing, no noisy neighbours on your card.

Dev-first interfaces

Prefer code? The CLI and Python SDK provision fleets without ever opening a dashboard.

Ready-made images

Start from PyTorch, TensorFlow or vLLM, bring a private registry image, or build from scratch. Anything that runs in Docker runs here.

Support from humans

Real people, not a bot loop. Higher tiers add onboarding, architecture review and guaranteed response times.

Security

Private by design

Your workloads, your data, your rules. Instances are isolated, access is yours alone, and nothing persists after you tear it down.

Talk to our team

Full environment control

Isolated instances with direct SSH, CLI and API access. No shared containers, root in your own box.

Data sovereignty

Delete models, data and workloads when you choose. Nothing persists without your command.

Region selection

Pin workloads to a region or a named facility when residency or latency dictates where the data can sit.

Scoped access

Separate keys per environment or team, each with its own spend cap. Revoke one without touching the rest.

FAQ

Common questions

What is OpenLink GPU Cloud?

A marketplace for GPU compute. Data centers, cloud operators and infrastructure partners list capacity; you search it, rent what fits, and release it when the job is done. You get bare-metal GPUs behind one account and one bill instead of a contract with each supplier.

How much does it cost?

Prices are set by supply and demand and move continuously. At the time of writing, RTX 4090 starts around $0.35/hr and H200 around $3.53/hr. The pricing page shows live numbers, and openlink search offers returns them programmatically.

How does billing work?

Per second, against prepaid credit, with a $5 minimum to start. On-demand holds the instance until you destroy it; interruptible capacity is cheaper but can be reclaimed; reserved locks a rate for a fixed term. Nothing accrues while nothing runs.

Which GPUs are available?

Blackwell (B200, RTX PRO 6000, RTX 5090), Hopper (H200, H100 SXM), Ada Lovelace (RTX 4090, L40S) and Ampere (A100, RTX 3090). Availability shifts with the market — query the current fleet from the CLI or the pricing page.

How do I deploy an instance?

Pick an offer, choose a Docker image, set disk and SSH, and launch. Three commands from the CLI, a few lines from the Python SDK, or a few clicks in the dashboard.

Is my data secure?

Instances are isolated with direct SSH access rather than shared containers. Data is removed when you destroy the instance, and access keys can be scoped per team or environment. For workloads under specific compliance regimes, talk to us about facility selection.

Developer tools

From zero to compute in seconds

Skip the quotas, skip the contracts, skip the chaos. Build, deploy and scale on your own timeline.

Get started Talk to sales