NVIDIA unveiled Vera CPU on March 23, 2026. The company describes it as the first processor purpose-built for the age of agentic AI and reinforcement learning, claiming 50% faster results and twice the efficiency of traditional rack-scale CPUs. The pitch is aimed at workloads where infrastructure around the model matters more, including task planning, tool execution, code running, and result validation.

The technical profile is aggressive. Vera uses 88 custom Olympus cores, up to 1.2 TB/s of LPDDR5X memory bandwidth, and NVIDIA's second-generation scalable coherency fabric. NVIDIA also introduced a Vera rack with 256 liquid-cooled CPUs, which it says can sustain more than 22,500 concurrent CPU environments. In the Vera Rubin NVL72 platform, Vera pairs with NVIDIA GPUs through NVLink-C2C with 1.8 TB/s of coherent bandwidth, which NVIDIA says is 7x PCIe Gen 6.

88 custom NVIDIA-designed Olympus cores
Up to 1.2 TB/s LPDDR5X memory bandwidth
More than 22,500 concurrent CPU environments in a Vera rack configuration, according to NVIDIA
Named partners include Alibaba Cloud, CoreWeave, Meta, Oracle Cloud Infrastructure, Dell, HPE, Lenovo, and Supermicro

The broader signal is that NVIDIA is no longer talking about the CPU as a passive companion to the GPU. Jensen Huang framed the CPU as a central part of how agentic systems scale, which fits the industry's shift from standalone model performance toward full rack-level optimization across memory, data movement, orchestration, networking, and security.

Performance and efficiency figures remain NVIDIA's own claims, so the market test will come when cloud providers and enterprises deploy Vera against existing x86 and Arm server fleets and compare total cost, migration complexity, and software compatibility. NVIDIA says Vera is already in full production and will be available from partners in the second half of 2026.

#vera

NVIDIA launches Vera CPU for agentic AI with 50% faster results and 2x efficiency claims