Nvidia Introduces Vera CPU, an '88-Core' Brain for AI Agents
Technology giant Nvidia has officially launched its latest central processing unit (CPU) named Vera. The chip was introduced during the Nvidia GTC (GPU Technology Conference) in Taiwan on Sunday (31/05/2026).
This CPU is specifically designed to handle Artificial Intelligence (AI) workloads, particularly AI capable of completing tasks autonomously, known as AI agents. Tasks that these AI agents can perform include data processing, reasoning, reinforcement learning, as well as executing code and using various tools automatically.
Vera is part of Nvidia’s next-generation AI platform, named NVIDIA Vera Rubin, which was unveiled by the company a year ago. The platform combines the Vera CPU with the Rubin Graphics Processing Unit (GPU) to manage training, inference, and large-scale AI agents within data centres.
“Vera is the first CPU designed for that future, and this chip is built to run AI agents at hyperscale with high performance, efficiency, and programmability,” said Jensen Huang.
Vera is designed using ARM architecture with custom CPU cores named Olympus. The chip features 88 CPU cores and supports 176 threads thanks to Spatial Multithreading technology. The CPU is paired with LPDDR5X memory, providing bandwidth of up to 1.2 TB per second and supporting RAM capacities of up to 1.5 TB.
These chips can be combined within a single rack, with up to 256 Vera CPUs providing a total of 22,528 cores and 45,056 threads. Nvidia claims this combination can accelerate various workloads, including code compilation, Python and Java processing, and database management.
In terms of performance, Nvidia claims that a single Vera CPU can complete various tasks up to 1.8 times faster than the x86-based CPUs currently widely used in data centres. According to Phoronix benchmarks, the Vera CPU can achieve peak performance across several common workloads used in AI agent operations.
In addition to serving as a standalone CPU for AI agents, data analytics, and reinforcement learning, Vera also acts as the primary processor within the Vera Rubin platform previously introduced by Nvidia. In the Vera Rubin NVL72 configuration, Nvidia combines 36 Vera CPUs and 72 Rubin GPUs, interconnected via NVLink-C2C with bandwidth of up to 1.8 TB per second.