Rent GB300 NVL72 racks.
72 Blackwell Ultra GPUs in one NVLink domain, per rack. Tell us how many racks you need and when. We check which operators can actually deliver and reply within 24 hours.
Why teams pick a full NVL72 rack
One 72-GPU NVLink domain
Every GPU in the rack talks to every other over NVLink at full speed. Large mixture-of-experts models and long-context inference can spread across 72 GPUs without touching the slower scale-out network.
More memory per GPU
288 GB of HBM3e per GPU and 37 TB of fast memory per rack. Bigger models and longer contexts fit per GPU, so you need fewer GPUs for the same job.
Built for reasoning inference
Microsoft measured about 1.1 million tokens per second of LLM inference from one ND GB300 v6 rack, roughly 27% more than GB200 (source).
Who rents out GB300 NVL72
Public clouds and neoclouds that say they offer GB300 NVL72, from their own pages. Being listed doesn't mean they have racks free for your dates. That's what we check.
| Provider | Status | What they say | Source |
|---|---|---|---|
| CoreWeave | Generally available | GB300 NVL72 instances through CoreWeave Kubernetes Service, launched 19 Aug 2025 in US-WEST-01A with more zones promised. Each rack also carries 18 BlueField-3 DPUs. | docs.coreweave.com Aug 2025 |
| Microsoft Azure | VM series | ND GB300 v6. Each VM is 4 GPUs and 2 Grace CPUs, 18 VMs per rack, 800 Gbps InfiniBand per GPU. Azure built the first large GB300 cluster for OpenAI. | learn.microsoft.com checked 9 Nov 2026 |
| AWS | Generally available | EC2 P6e-GB300 UltraServers, GA 2 Dec 2025. AWS says to contact your AWS sales representative to get started. | aws.amazon.com Dec 2025 |
| Oracle (OCI) | Superclusters | OCI Supercluster configurations built on GB300 NVL72 racks, listed as orderable in March 2025. | blogs.oracle.com Mar 2025 |
| Nebius | Through sales | Says GB300 NVL72 is deployed in production on Nebius AI Cloud and sold through its sales team, configured per workload. | nebius.com checked 9 Nov 2026 |
| Lambda | Superclusters | Single-tenant GB300 NVL72 superclusters on InfiniBand, listed on its site. | lambda.ai checked 8 Nov 2026 |
| Crusoe | Through sales | Sells GB300 NVL72 and HGX B300 through its sales team. | crusoe.ai checked 8 Nov 2026 |
How a quote works
- You send the basics.Configuration, rack count, term and start date. A sentence about the workload helps.
- We ask operators that fit.Neoclouds, data center operators and owners with racks coming free, not just the big names above.
- You get options within 24 hours.Who can deliver, when, on which fabric, and on what terms. You contract with the provider directly.