Nvidia Ships First Vera CPUs to SpaceX; Musk Praises Chip

Nvidia delivered early units of its Vera CPU directly to SpaceX and several cloud and AI customers as part of an initial rollout. The company’s hyperscale division hand-delivered the first chips to begin on-site testing.

Nvidia introduced Vera as its first processor designed specifically for agentic AI workloads at its March 2026 GPU Technology Conference, where it unveiled the Rubin platform. The vendor said the initial silicon is now in testing at early customers while broader commercial deployment starts.

Vera uses 88 custom Olympus cores and supports up to 1.2 TB/s of memory bandwidth with LPDDR5X memory. Nvidia stated the chip can run agentic sandbox workloads up to 50% faster than competing rack-scale CPUs and offers roughly double the energy efficiency, according to the company’s performance claims.

The company outlined a 256-CPU rack configuration that it says can sustain more than 22,500 concurrent agent environments at full performance. Nvidia positioned Vera to handle workloads that coordinate many smaller models and run large numbers of independent agent instances.

Recipients of the first units include SpaceX, OpenAI, Anthropic and Oracle Cloud Infrastructure. Nvidia also named Alibaba Cloud, ByteDance, Meta, CoreWeave, Lambda and Nscale among early adopters.

SpaceX has folded xAI into the company under the SpaceXAI banner. Nvidia said the Vera unit is already running workloads tied to the Colossus 1 and Colossus 2 supercomputers in Memphis. Elon Musk reposted Nvidia’s announcement on X and added that the chip “lived up to its name.”

Nvidia framed the deliveries as the start of a wider commercial deployment that places Vera-based server racks alongside GPUs and other tools for agent orchestration on the Rubin platform. AMD and other chipmakers are developing processors aimed at similar workloads. The timeline for customers to move from testing to full production will shape whether Vera gains meaningful CPU market share.