CoreWeave adds NVIDIA Vera CPU to its cloud portfolio

By Splitfeed EditorialLast updated Sep 30, 2026, 15:30 UTC

On September 30, 2026, at its Fully Connected event in San Francisco, CoreWeave announced it would offer NVIDIA's Vera CPU as a bare-metal service at rack scale. NVIDIA describes the Vera as the first CPU designed specifically for AI agent workloads. The announcement came one day after Los Alamos National Laboratory received a Vera CPU unit, on September 29. AI coding company Cognition was named as an early customer for the Vera Rubin NVL72 system. CoreWeave stock rose following the news, according to Barron's. NVIDIA's blog framed the partnership as the product of nearly a decade of co-engineering between the two companies.

The announcement matters because it marks a concrete step toward infrastructure built not for model training but for agentic AI, meaning systems where AI models take sequences of actions with some degree of autonomy. Offering Vera at rack scale as bare metal signals that CoreWeave is targeting customers who want dedicated, high-performance compute for running AI agents in production, not just shared cloud resources. The involvement of Los Alamos National Laboratory, a major scientific and defense research institution, suggests the hardware is already moving beyond pilot announcements into real deployments. What remains genuinely in dispute is whether purpose-built CPU infrastructure for AI agents represents a meaningful architectural shift or whether it is primarily a marketing distinction layered onto existing compute trends.

None of the three camps, Pro-AI, Anti-AI, or Middle Ground, have published reactions to this story yet. The Pro-AI camp would typically welcome this as evidence that the AI infrastructure stack is maturing rapidly, with specialized hardware now targeting agentic use cases as a distinct category. The Anti-AI camp would likely raise questions about the concentration of AI-optimized compute in a small number of commercial providers, and about what large-scale agentic AI deployment means for accountability and oversight. The Middle Ground camp would probably focus on whether demand for agentic AI infrastructure is real and sustainable, or whether adoption by early customers like Cognition and institutions like Los Alamos is too narrow to draw broad conclusions from.

The clearest signal to watch for next is how quickly additional customers are named for the Vera Rubin NVL72 system and whether CoreWeave provides any pricing or availability timeline for the bare-metal Vera service. Broader enterprise uptake, or the lack of it, would help settle the question of whether agentic AI infrastructure is an emerging category or an early-stage niche.