The AI cloud provider linked a group of high performance hardware racks on its platform to increase the capacity for agentic AI workloads. Each Vera Rubin NVL72 rack incorporates 72 Rubin GPUs and 36 Vera CPUs, using Spectrum-X networking to provide 1.6 Tb/s of connectivity per GPU. This system fabric is designed to scale to approximately 128,000 GPUs per rail.

CoreWeave also launched a low cost Archive tier for AI Object Storage and a LOTA storage accelerator. The accelerator provides 7 GB/s throughput per GPU and an 8x reduction in read latency versus traditional storage clusters. The company says it is the first AI cloud provider to validate a single rack Vera Rubin NVL72 before moving to multi rack deployments.