19/05/2026
Nvidia just hand-delivered the first Vera CPUs.
Not shipped. Not flown commercial. Hand-delivered, by Ian Buck, Nvidia's VP of Hyperscale and HPC, personally walking units into customer offices.
The stops:
β Friday: Anthropic (SF), OpenAI (Mission Bay), SpaceXAI (Palo Alto)
β Monday: Oracle Cloud Infrastructure (Santa Clara)
Vera is Nvidia's first custom Arm-based server CPU, purpose-built for agentic AI workloads and designed to pair tightly with Rubin GPUs over NVLink. It replaces Grace and marks the CPU half of the Vera Rubin platform, moving from announcement to production.
A few things stand out:
1. The customer list. Three frontier labs and one hyperscaler, each of them a strategic anchor for the agentic AI buildout.
2. The white-glove choreography. When a VP personally couriers silicon, that's not logistics, that's positioning.
3. The timing. Right before Nvidia's earnings, with Vera Rubin systems expected to ship at scale in 2H 2026.
The AI infrastructure story in 2026 isn't going to be about GPUs alone. It's going to be about the full stack CPU, GPU, networking, and memory, moving in lockstep. Vera is the piece most people weren't watching.
Now worth watching.