1 hour ago
27
CoreWeave just became the first company on the planet to power up Nvidia’s Vera Rubin NVL72 rack-scale system. The AI cloud provider completed operational validation of the hardware on June 1, 2026, one day after Dell Technologies dropped off what amounts to the most powerful commercially deployed AI inference machine ever built.
Each rack packs 72 Rubin GPUs and 36 Vera CPUs, connected by NVLink 6 fabric running at 260 TB/s. Each rack is also capable of 3.6 exaFLOPS of NVFP4 inference capability.
A tenfold jump that matters
The Vera Rubin NVL72 delivers up to 10x better token throughput per megawatt on complex workloads like DeepSeek R1 compared to Nvidia’s previous Blackwell platform. In practical terms, it means dramatically fewer GPUs are needed to handle equivalent workloads, which translates directly into lower costs per million tokens for large-scale inference applications.
Dell’s integration speed deserves its own mention. The team took the system from physical delivery to production readiness in under 6.5 hours.
CoreWeave’s strategic positioning
CoreWeave and Dell have been building on a partnership that spans prior NVL system generations. CoreWeave’s stock spiked roughly 14% following the announcement.
What comes next
Nvidia expects the Vera Rubin platform to enter full production by July 2026. Deployments are planned across OpenAI, Google Cloud, and Microsoft Azure.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·