Overview
- CoreWeave announced on Monday, June 1 that it completed NVIDIA’s system-level bring-up and validation of the Vera Rubin NVL72 rack, making it the first AI cloud provider to declare the platform production-ready.
- NVIDIA’s Vera Rubin is a liquid-cooled rack that houses 72 Rubin GPUs and 36 Vera CPUs and is claimed to deliver big gains in inference efficiency and lower token costs versus prior architectures.
- To run the racks at scale, CoreWeave built rack-level controls and software such as Valvey, Racky and a Rack Lifecycle Controller to manage liquid cooling, power, networking and multi-tenant orchestration.
- The technical milestone arrived alongside major commercial moves: a reported near-$100 billion contract backlog, FY2026 revenue guidance of $12–$13 billion, a closed $3.1 billion infrastructure loan facility and an investment in inference optimizer Tensormesh.
- Practical limits remain: CoreWeave and others say broader deployments depend on data-center liquid-cooling readiness and orchestration integration, and the company’s early-adopter track record follows its pivot from crypto infrastructure to large-scale AI cloud services.