CoreWeave Deploys Nvidia Vera Rubin NVL72 Clusters for AI Clouds

CoreWeave is the first AI cloud provider to validate and deploy Nvidia’s Vera Rubin NVL72 architecture to improve performance for complex, agentic AI workloads. This deployment changes high-performance computing capacity by using networking and storage tiers to support scale-out clusters.

"CoreWeave was the first AI cloud provider to validate and bring up a Vera Rubin NVL72, demonstrating that this advanced rack-scale architecture could operate as a reliable, high-performance cloud service," said Chen Goldberg, executive vice president of product & engineering at CoreWeave.

Technical Architecture and Scale

Each Vera Rubin NVL72 rack is built for high-density computing, housing 72 Rubin GPUs and 36 Vera CPUs. The architecture integrates Nvidia NVLink 6 technology, ConnectX-9 SuperNICs, and BlueField-4 DPUs to manage communication between components.

To manage these clusters, the system uses Nvidia Spectrum-X Ethernet networking. This configuration allows for the connection of hundreds of GPUs into a single scale-out cluster, supporting a theoretical limit of approximately 128,000 GPUs per rail.

New Storage Capabilities

Alongside the hardware rollout, CoreWeave has introduced two specific storage features that are available now:

• Cross-region write acceleration: This feature synchronizes data footprints across geographically dispersed data centers.
• Archive storage tier: This cost-management feature eliminates retrieval, early deletion, and reading fees for stored data.

These enhancements reduce latency for AI models that require constant data access across regions. Cécile Robert-Michon, director of internal infrastructure at Cohere, noted the utility of this approach, stating, "CoreWeave AI Object Storage gives us a unified dataset footprint across regions with reads cached locally, so nothing waits on the network."

Financial Context

CoreWeave (CRWV) stock saw a modest increase on September 18, 2026, closing at $81.36 per share, up 1.85%. The company's focus on deploying Nvidia hardware as a cloud service is part of an industry strategy to provide infrastructure for firms developing generative and agentic AI models, which require significant compute power and rapid data throughput to function effectively.