NVIDIA has begun shipping its first custom CPU designed specifically for agentic AI, the Vera, to a lineup of early adopters that includes SpaceX, OpenAI, Anthropic, and Oracle Cloud Infrastructure. The chip, built around 88 custom Olympus cores and backed by up to 1.2 TB/s of memory bandwidth using LPDDR5X, targets the rapidly growing demand for systems that can run thousands of autonomous AI agents at once.
What Vera brings to the rack
The Vera CPU is part of the broader Rubin platform that NVIDIA unveiled at GTC in March 2026. Each processor handles agentic sandbox workloads — environments where AI agents operate and learn — up to 50% faster than competing rack-scale CPUs, according to the company, while doubling efficiency. A single rack configuration with 256 Vera CPUs can sustain more than 22,500 concurrent agent environments at full performance, a figure that underscores the scale NVIDIA is aiming for.
CEO Jensen Huang described agent-based services as NVIDIA's next multi-trillion-dollar opportunity, putting Vera at the center of that bet. The chip is the first NVIDIA-designed CPU, marking a shift from relying on third-party processors in its platforms.
Elon Musk and the SpaceXAI connection
Elon Musk, whose SpaceX is now home to the xAI team under the SpaceXAI brand, publicly cheered the Vera launch on X. He reposted NVIDIA's announcement and added a pun: "Vera nice, Vera nice..." The SpaceXAI division operates the Colossus 1 and Colossus 2 supercomputers in Memphis, which will likely tap into Vera's capacity for running large numbers of agents.
Musk's companies have been among the most aggressive in deploying large-scale AI infrastructure. The integration of xAI into SpaceX means that Vera will power workloads spanning space exploration, autonomous systems, and general AI research under one roof.
Beyond the early adopters
Other companies that have received Vera units include Alibaba Cloud, ByteDance, Meta, CoreWeave, Lambda, and Nscale. The range of adopters — from cloud giants to specialized GPU infrastructure providers — suggests that agentic AI workloads are already spreading beyond a handful of labs.
Competitors like AMD are working on chips aimed at similar agentic AI workloads, but NVIDIA is first to market with a purpose-built CPU. Whether Vera's performance claims hold up in real-world deployments will determine how fast the market shifts.
The Rubin platform, with Vera at its core, is expected to roll out more broadly in the coming months. NVIDIA has not disclosed pricing or a general availability date, but early shipments to influential customers indicate that production is underway. The big question left unanswered: how quickly will the rest of the industry follow SpaceX and Oracle into agentic AI at scale?
