CoreWeave brings up first NVIDIA Vera Rubin NVL72 — 72 Rubin GPUs, 36 Vera CPUs, 3.6 exaFLOPS validated in 147 hours after Dell delivery on May 31
CoreWeave became the first AI cloud provider to bring up and validate NVIDIA's Vera Rubin NVL72 platform on June 1, 2026. The liquid-cooled rack — 72 Rubin GPUs paired with 36 Vera CPUs delivering 3.6 exaFLOPS at FP4 — was delivered by Dell on May 31 and cleared NVIDIA's L11 diagnostics plus a 147-hour validation suite before CoreWeave declared production readiness. CRWV stock closed up 13.96% at $124.82 on the news.
The interesting piece is the validation cadence. Nvidia announced Vera Rubin general availability at GTC 2026 in March; the platform was supposed to ramp through Q2 with full production targeted for H2. CoreWeave compressed the timeline by partnering with Dell on the first rack delivery and running the L11 diagnostic plus a 147-hour soak in parallel with the GA announcement — the bring-up was effectively rehearsed in advance so that the validation receipt could be published the moment Nvidia opened the door. The Dell + CoreWeave choreography is the operational story; the chip is the marketing story.
The technical detail worth holding is the memory architecture. The NVL72 rack treats its 72 Rubin GPUs as a single memory domain over 7th-generation NVLink, with HBM4 on every socket — the design target is frontier-model inference at trillion-parameter scale where the entire model state plus KV cache must remain coherent across the rack. CoreWeave also disclosed two pieces of in-house software that ship with the rack: Valvey, a software-defined per-rack valve assembly that makes liquid cooling a programmable control surface, and the Rack Lifecycle Controller, a Kubernetes-native orchestrator that treats the full NVL72 as one programmable entity. The differentiation against AWS, Azure, and Oracle's coming Vera Rubin deployments is the orchestration layer rather than the silicon.
The Meta angle anchors why the timing matters. The $21B CoreWeave–Meta agreement signed in March 2026 explicitly named Vera Rubin as part of the dedicated capacity; first-rack validation is the gate that releases the deployment schedule on that contract. The June 1 bring-up therefore reads as both a CoreWeave milestone and a Nvidia supply-chain demonstration — Vera CPUs, BlueField-4 STX storage, Spectrum-X photonics, and the Rubin GPUs themselves all had to clear the same validation pass simultaneously across the 350+ factories in 30 countries that Nvidia disclosed as supply partners during its May 31 production-ramp press release.
CoreWeave — CoreWeave completes industry-first bring-up and validation of NVIDIA Vera Rubin NVL72 → · SiliconANGLE — Nvidia ramps up production of Vera Rubin, the foundation of the next generation of AI factories → · TechTimes — CoreWeave beats all rivals to NVIDIA Vera Rubin NVL72: CRWV stock surges 14% →