NVIDIA's Vera Rubin Era Goes Live: CoreWeave Brings Up First Multi-Rack Vera Rubin NVL72 Cluster
On September 16, 2026, CoreWeave announced the bring-up of multi-rack NVIDIA Vera Rubin NVL72 on its cloud — putting "hundreds of NVIDIA Rubin GPUs into a single scale-out cluster for agentic AI." CoreWeave claims it was "the first AI cloud provider to validate and bring up a Vera Rubin NVL72," meaning Nvidia's next-generation datacenter platform has moved from spec sheets to operating production cloud infrastructure.
What a rack looks like. A single Vera Rubin NVL72 rack pairs 72 Rubin GPUs with 36 Vera CPUs, NVIDIA NVLink 6, ConnectX-9 SuperNICs, and BlueField-4 DPUs. Multi-rack configurations unify racks via Spectrum-X Ethernet into one scale-out cluster — every Rubin GPU gets two ConnectX-9 SuperNICs for 1.6 Tb/s of scale-out connectivity per GPU, supporting "roughly 128,000 GPUs per rail in a non-blocking fabric," with a modular topology so racks can be added without redesigning the fabric.
"With multi-rack Vera Rubin, we are connecting hundreds of Rubin GPUs as a single scale-out cluster. For customers building agentic AI, that means greater scale, faster iteration, and higher productivity as models and agents continuously learn and improve." — Chen Goldberg, EVP product & engineering, CoreWeave
Why it matters. This is the first concrete operational milestone in the Vera Rubin ramp — the successor to Blackwell that the entire US compute buildout (OpenAI's Ohio campus, Anthropic's datacenter financing, CoreWeave's fleet) is counting on. CoreWeave is positioning rack-scale engineering (automated rack lifecycle control, full-system validation, storage co-designed with the GPUs — including cross-region write acceleration and a new Archive tier for checkpoint data) as its moat. Expect competing bring-up announcements from other neoclouds and hyperscalers as Rubin volume ramps through late 2026.