CoreWeave became the first AI cloud provider to operationalize Nvidia’s Vera Rubin NVL72 system, connecting hundreds of next-generation GPUs in a production cluster for demanding AI workloads. Its stock, CRWV, surged roughly 14% and trading volume nearly doubled its three-month average after the announcement. Dell delivered the NVL72 hardware on May 31, and CoreWeave had it running by June 1, a turnaround of about 24 hours. Each rack contains 72 Rubin GPUs and 36 Vera CPUs linked by NVLink 6, with 260 TB/s of bandwidth and up to 3.6 exaFLOPS of NVFP4 inference capacity. CoreWeave says Rubin provides 10 times the inference throughput per megawatt of the previous Blackwell generation, while reportedly requiring half as many GPUs for equivalent workloads and reducing the cost per million tokens to roughly one-tenth of Blackwell. The architecture can scale to thousands of GPUs through Spectrum-X Ethernet or Quantum-X800 InfiniBand. CoreWeave also introduced Valvey for real-time monitoring of liquid-cooling performance, Racky for unified rack control, and LOTA storage capable of data transport speeds of up to 7 GB/s per GPU. The deployment reinforces CoreWeave’s strategy of offering Nvidia’s newest systems ahead of hyperscalers, creating a period of exclusivity for customers willing to pay premium prices. Full production availability of Rubin clusters is expected later in 2026, with major hyperscalers also planning adoption. CoreWeave’s 10-times efficiency and cost claims come from its own benchmarks, while its claim of four-times faster training in certain scenarios remains to be independently assessed; validation could affect the pace of next-generation AI development and demand for Rubin clusters.