CoreWeave unveils first Nvidia Vera Rubin NVL72 cluster

CoreWeave deployed the first Nvidia Vera Rubin NVL72 cluster, achieving up to 10x Blackwell's efficiency and boosting AI storage. Shares rose up to 14% after the news.

CoreWeave has launched the first multi-rack Nvidia Vera Rubin NVL72 system in production, connecting hundreds of next-generation GPUs in a cluster tailored for high-demand AI workloads. According to the company, this platform delivers up to 10 times greater inference efficiency per megawatt compared to the Blackwell generation. It also halves the number of GPUs required for similar tasks and reduces the cost per million tokens to just one-tenth. The deployment was completed within 24 hours of receiving the hardware. It features enhancements to AI object storage, including accelerated cross-region writes and a new Archive layer. Following the announcement, CoreWeave (CRWV) shares surged by up to 14%, while Nvidia (NVDA) saw a 1% increase. This milestone strengthens CoreWeave’s position in the competitive AI infrastructure market and demonstrates strong demand for Rubin technology among cloud providers. However, risks remain if clients reduce training expenditures or if major providers delay investments.