CoreWeave Launches Forge and Vera Rubin NVL72 at First Fully Connected Event
CoreWeave used its first Fully Connected conference in San Francisco to launch CoreWeave Forge, a platform for building, running and improving AI models and agents, and to announce availability of NVIDIA's Vera Rubin NVL72 system. According to the company, more than 3,000 AI practitioners attended.
- CoreWeave held its first Fully Connected conference in San Francisco with more than 3,000 attendees.
- CoreWeave Forge is a single place to build, run and improve AI models and agents.
- NVIDIA Vera Rubin NVL72 is available, with Cognition the first customer running production workloads on it.
- NVIDIA Vera CPU supports more than 11,000 concurrent environments per rack, and sandboxes start more than three times faster than on x86, per CoreWeave's testing.
- The new CoreWeave Partner Network includes CrowdStrike, VAST Data, ClickHouse and Reflection.
Forge brings the AI loop into one place
CoreWeave Forge is described by the company as "one place to build, run, and improve AI models and agents." The idea behind it, according to CoreWeave, is that what a team learns from real AI use feeds the next version and makes it better.
Chen Goldberg and Corey Sanders walked through the steps of what the company calls the AI loop live on stage to demonstrate Forge. CoreWeave is inviting interested teams to learn more and sign up on its site.
New NVIDIA hardware: Vera Rubin NVL72 and Vera CPU
CoreWeave announced the availability of NVIDIA Vera Rubin NVL72. Cognition is the first customer anywhere running production workloads on NVIDIA's newest system, according to CoreWeave. The company says the deployment was up and running within days and delivered more token throughput than the previous generation.
CoreWeave also now offers NVIDIA Vera CPU, which it calls the first CPU built for AI agents. These processors handle the work around each agent step, from sandboxes to tool calls. The company says one rack supports more than 11,000 concurrent environments, and that in its own testing sandboxes start more than three times faster than on an x86 CPU.
Healthcare customer and a new Partner Network
CoreWeave said Ennoble Care, a home-based care provider that serves about 50,000 patients a year, selected it to run its clinical AI. The company points to dedicated capacity, security built in from the silicon up, and engineers who answer the phone as the elements meant to help bring AI into everyday care.
The company also unveiled the CoreWeave Partner Network, which collects integrations from partners including CrowdStrike, VAST Data, ClickHouse and Reflection in one place. According to CoreWeave, each integration has been tested on its infrastructure under production load, so customers can scale up without rebuilding what already works.
What it means for AI teams
The conference featured customers such as Caterpillar, Capital One and Cognition. CoreWeave says what they share is that AI must prove itself on the job every day, at a scale that keeps growing. The company argues this calls for a purpose-built AI cloud, since systems designed for earlier eras of computing were not built for these demands.
The announcements show CoreWeave positioning itself as a full AI platform rather than only a supplier of infrastructure, combining a build-and-run environment, new NVIDIA hardware and a set of pre-tested partner integrations. The performance figures cited come from the company and its own testing, and the source does not give pricing or broader availability details.
