CoreWeave은 다중 랙 Nvidia Vera Rubin NVL72 클러스터를 공식 가동했으며, 이는 에이전트 AI 워크로드를 위한 차세대 아키텍처의 첫 대규모 배포를 의미합니다.

CoreWeave Inc. (CRWV) announced Wednesday that it has brought a multi-rack Nvidia Vera Rubin NVL72 cluster online via CoreWeave Cloud, linking hundreds of Nvidia Rubin graphics processing units (GPUs) into a single system optimized for agentic AI workloads. The deployment enables training and inference tasks to span hundreds of Rubin GPUs, delivering the capacity required for larger models, more demanding inference, and large-scale reinforcement learning.

A single Nvidia Vera Rubin NVL72 rack integrates 72 Rubin GPUs with 36 Vera CPUs, alongside Nvidia networking and data-processing technology. Multiple racks are interconnected via Nvidia Spectrum-X Ethernet to function as a unified scale-out cluster. According to Chen Goldberg, Executive Vice President of Product and Engineering, CoreWeave was the first AI cloud provider to validate and deploy the Vera Rubin NVL72 architecture.

“With multi-rack Vera Rubin, we are connecting hundreds of Rubin GPUs as a single scale-out cluster,” Goldberg said, adding that the system gives customers building agentic AI greater scale and faster iteration. CoreWeave’s infrastructure automates rack setup, firmware updates, validation, power management, and cooling. Full racks undergo comprehensive testing before being moved into production.

Alongside the compute expansion, CoreWeave introduced new storage capabilities designed to keep data closer to computing resources across regions while lowering long-term storage costs. The company launched cross-region write acceleration for its AI Object Storage platform, allowing AI workloads to write data in their local region while CoreWeave asynchronously copies it to another region in the background. This capability minimizes pauses for customers running distributed training workloads. Additionally, CoreWeave introduced an Archive tier—a lower-cost storage option for long-term data retention that carries no retrieval fee, early-deletion fee, or fee for reading data from the Archive tier.

The company also highlighted its Local Object Transport Accelerator (LOTA), which caches data closer to AI workloads. CoreWeave reports that LOTA reduces read latency by eight times compared to reading from a traditional storage cluster.

Retail investor sentiment surrounding CRWV shares remains in bearish territory on Stocktwits as of September 16, 2026. Year-to-date, CoreWeave stock has risen approximately 2%. By comparison, the ARK Innovation ETF (ARKK), which holds CRWV, has gained over 6%, while the iShares AI Innovation and Tech Active ETF (BAI) has advanced more than 25% during the same period.