Nvidia Releases Vera CPU: Designed for Agent AI, Alibaba, ByteDance, Meta and Others Plan Deployment

Nvidia officially released the Vera CPU at the GTC conference, the first processor designed specifically for Agent AI and reinforcement learning. As AI expands from generation and inference to autonomous action, workflows like task planning and tool calling have higher CPU requirements. Jensen Huang said: 'The CPU is no longer just an auxiliary model; it is driving the model.'

Vera uses 88 self-developed Olympus cores, each capable of executing two tasks simultaneously via Spatial Multithreading, suitable for parallel operations in multi-tenant AI factories. It uses second-generation low-power LPDDR5X memory with a bandwidth of 1.2 TB/s, double that of general CPUs and half the power consumption.

Cloud service providers like Alibaba, ByteDance, and Meta have planned deployment. 19 hardware vendors including Dell and HPE are adapting it. Cursor will use it to improve programming Agent performance, and Redpanda tests show up to 5.5x lower latency. Several U.S. labs also plan deployment. Vera is in mass production and will ship in the second half of this year.