CoreWeave (CRWV) shares climbed roughly 3% in premarket trading on Wednesday after the company announced it had deployed a multi-rack Nvidia Vera Rubin NVL72 cluster on CoreWeave Cloud, becoming the first AI cloud provider to validate and bring up the architecture at multi-rack scale.
The new setup connects hundreds of Nvidia Rubin GPUs into a single scale-out cluster, allowing customers to run training and inference workloads across multiple racks. A single Vera Rubin NVL72 rack combines 72 Rubin GPUs with 36 Vera CPUs, Nvidia NVLink 6, ConnectX-9 SuperNICs, and BlueField-4 DPUs. Multiple racks are stitched together using Nvidia Spectrum-X Ethernet, with ConnectX-9 SuperNICs delivering 1.6 Tb/s of scale-out connectivity per GPU and supporting roughly 128,000 GPUs per rail in a non-blocking fabric.
Chen Goldberg, EVP of product and engineering at CoreWeave, said the multi-rack configuration gives customers building agentic AI “greater scale, faster iteration, and higher productivity as models and agents continuously learn and improve.” The company also said its infrastructure automates rack setup, firmware updates, validation, power and cooling processes, and tests complete racks before they enter production.
Alongside the compute expansion, CoreWeave launched two new AI Object Storage features. Cross-region write acceleration lets workloads write data locally while the platform replicates it to a remote region in the background. A new Archive tier offers lower-cost long-term data retention with no retrieval, early-deletion, or reading fees. CoreWeave’s Local Object Transport Accelerator (LOTA) delivers reads at local NVMe speeds and up to 7 GB/s of throughput per GPU, reducing latency by up to 8x compared to traditional storage clusters.
Nvidia (NVDA) edged up about 1% on the news. However, CoreWeave’s stock later pared some gains after the Federal Reserve raised interest rates to 4% in a unanimous decision and signalled further rate increases. Bernstein maintained its Sell rating on CoreWeave while raising its price target to $74, citing risks from reduced AI training spending and concentration in rural data-centre markets. The company also received a boost from a definitive 15-year anchor lease agreement with Blockfusion USA’s subsidiary North East Data for capacity at its Niagara Falls, New York campus.
