CoreWeave Plans Nvidia Vera CPU Service for AI Agents

CoreWeave (NASDAQ: CRWV) Inc. plans to offer Nvidia Corp.’s Vera processors as standalone computing capacity for artificial intelligence agents, expanding its cloud services for the work surrounding AI models.

The company said Sept. 30 that its configuration would support more than 11,000 concurrent computing environments per rack, using 128 Vera CPUs. The service is “coming soon,” according to an accompanying company blog post, which did not specify a launch date or pricing.

The offering addresses a growing part of AI infrastructure: running the software tools, code and data-processing tasks that agents use to complete assignments. While graphics processors handle model computation, CPUs execute much of that surrounding work. Nvidia identified those requirements as a focus when it launched Vera in March.

CoreWeave said its tests showed agent sandboxes—isolated environments for executing code—started more than three times as fast on Vera as on an alternative x86 processor. The post did not identify the comparison chip or provide enough benchmark detail to assess how broadly the result applies. Vera will join AMD EPYC and Intel Xeon in CoreWeave’s CPU portfolio, allowing customers to expand CPU capacity independently of their GPU fleets.