CoreWeave Deploys Vera Rubin, Integrates Training and Inference

Continuous, agentic AI workloads change the cost and operational calculus for infrastructure teams because always-on sessions and million-token contexts make token-level efficiency and rack-scale reliability first-order concerns. Reporting from CoreWeave's blog and coverage by SiliconANGLE and TechZine shows how CoreWeave validated NVIDIA Vera Rubin NVL72 at rack scale and built an integrated stack that combines training, continuous inference, observability, and autonomous improvement. CoreWeave's blog states the company was the first cloud provider to bring up and validate NVIDIA "Vera Rubin NVL72". SiliconANGLE quotes Chen Goldberg, CoreWeave Executive VP of Product and Engineering, on token economics and full-stack coordination. TechZine reports CoreWeave is offering an integrated platform that includes Serverless RL, CoreWeave Inference, W&B Weave observability, and W&B Skills, and it attributes claims of training being about 1.4x faster and up to 40% cheaper to CoreWeave as reported in that coverage.