Rent NVIDIA Vera Rubin NVL72 racks.
The first racks are live and the next ones are being reserved now. Tell us how many you need and when. We find the operators who can deliver and reply within 24 hours.
Preliminary figures from NVIDIA. The 10x figure is NVIDIA's claim against GB200 NVL72.
Built for the token economy
Ten times the tokens from the same megawatt
Power is the limit on every AI build now. NVIDIA says Vera Rubin delivers up to 10x more tokens per megawatt than GB200 NVL72, at one-tenth the cost per million tokens (NVIDIA). CoreWeave measured roughly 10x tokens per megawatt on DeepSeek R1 in its own test (CoreWeave).
4.8x in production
Cognition, the first customer running production work on the rack, reported 4.8x total token throughput against a GB200 NVL72 baseline (CoreWeave).
NVLink 6, doubled
260 TB/s across the rack, 3.6 TB/s per GPU. All 72 GPUs behave as one for large mixture-of-experts models (NVIDIA).
Train with a quarter of the GPUs
NVIDIA says mixture-of-experts models train on one-fourth as many GPUs as the previous platform (NVIDIA). Fewer racks for the same model changes the size of the hall you need, and the size of the contract.
Who has Vera Rubin NVL72
Where each cloud stands, from public announcements. Listed doesn't mean available to you: most of this capacity is spoken for before it's racked. We find what's actually open.
| Provider | Status | What's public | Source |
|---|---|---|---|
| CoreWeave | Customers live | Says Vera Rubin NVL72 is available on its cloud. Its first customer, Cognition, has run production workloads on a cluster stood up in early September 2026. | coreweave.com 30 Sep 2026 |
| Google Cloud | First instances | NVIDIA says the first A5X instance on Vera Rubin is up and running for a London startup. | blogs.nvidia.com 21 Jul 2026 |
| Microsoft Azure | Racks running | Named by NVIDIA among partners with racks running. Microsoft and Mistral plan European infrastructure with tens of thousands of GPUs. | blogs.nvidia.com 21 Jul 2026 |
| Oracle (OCI) | Racks running | Named by NVIDIA among partners with racks running, and an early adopter of co-packaged optics for the network between racks. | blogs.nvidia.com 21 Jul 2026 |
| Nebius | First system received | Received its first Vera Rubin NVL72 at its Finland site and plans availability in Europe and the US. | blogs.nvidia.com 21 Jul 2026 |
| AWS | Announced | Named by NVIDIA at launch among the clouds expected to deploy Vera Rubin instances in 2026. | nvidianews.nvidia.com 5 Jan 2026 |
| Lambda | Announced | Named at launch as an NVIDIA Cloud Partner expected to deploy in 2026. | nvidianews.nvidia.com 5 Jan 2026 |
| Nscale | Announced | Named at launch as an NVIDIA Cloud Partner expected to deploy in 2026. | nvidianews.nvidia.com 5 Jan 2026 |
How it works
- You tell us the shape of the deal.Rack count, term, start date and region. A line about the workload helps.
- We go to the operators.Clouds, neoclouds and data center operators with Vera Rubin allocations, including ones that don't advertise.
- You get real options within 24 hours.Who can deliver, when, at which site, and on what terms. You contract with the provider directly.