Cognition has become the first customer in the world to run production workloads on Nvidia's newest AI hardware, utilizing the platform to scale its Devin software engineering agents. CoreWeave has made the Vera Rubin NVL72 systems available for production use with Spectrum-X networking, enabling Cognition to achieve a 4.8x increase in token throughput over the GB200 NVL72 baseline. CoreWeave also released Forge, a unified set of training, evaluation and deployment tools designed to streamline model and agent improvements across different hardware vendors.

The deployment follows CoreWeave's acquisition of Ennoble Care as a client to run clinical AI inference using RTX PRO 6000 Blackwell nodes. The cloud provider also launched a Partner Network featuring validated production integrations with CrowdStrike, VAST Data, ClickHouse, Reflection, Exa and You .com. Market demand for the Rubin systems is currently extending into 2027.