Nvidia’s Vera CPU aims to redefine AI servers, challenging AMD and Intel

Nvidia unveils Vera, its first-in-house data-center CPU designed to accelerate AI workloads, prioritizing single-core speed, memory bandwidth, and latency to better feed AI agents. Delivered to clients like OpenAI, Anthropic, and SpaceX, Vera can be bought standalone or paired with Nvidia GPUs in systems such as Vera Rubin, and can be configured in large 256-chip liquid-cooled racks. The chip draws 250–450W and supports up to 1.5 TB memory per chip. Nvidia positions Vera as part of a vertically integrated AI compute stack to outpace incumbents, while analysts estimate a potential $200B server-CPU market and note uncertain adoption hurdles as AMD/Intel retain strong ecosystems.