NVIDIA’s CEO, Jensen Huang, recently made a significant announcement: the company’s new Vera CPU will soon enter the market as an independent infrastructure product, stepping out from its former supporting role alongside GPUs. This move signifies NVIDIA’s first foray into commercializing a processor capable of independently powering a full computing stack, directly challenging the dominance of AMD’s EPYC and Intel’s Xeon in the data center market. Huang underscored the groundbreaking nature of the Vera CPU, sharing that early partner CoreWeave has already outlined plans for its deployment, though official design orders remain pending full disclosure.

Specifically engineered for agentic inference, the Vera CPU boasts 88 custom Olympus cores, supports 176 threads, and achieves a staggering 1.8TB/s bandwidth connectivity via the NVLink-C2C bus. This performance doubles that of the previous-generation Grace CPU. The product is slated to begin shipping to initial customers, including Microsoft, in the latter half of 2026. It will serve as the central component of the Vera Rubin superchip combination, creating a next-generation AI computing platform when paired with the Rubin GPU.