Musk Puts Even CPU on the Line Against Nvidia: SpaceX's AI Department Deploys Vera to Run Agent

The latest Beating AI News reveals that SpaceXAI will deploy the NVIDIA Vera CPU to power the next-generation AI Agent. It is not intended to replace GPUs for training large models but rather to handle activities between model calls, including tool invocation, code execution, data handling, task orchestration, and simulation. SpaceXAI will also continue to expand Grok's AI infrastructure with Vera Rubin.

As AI Agents become more complex, CPUs are becoming more involved. After model generation tool invocations, the actual execution of code, data retrieval, and other tasks usually fall on the CPU. If the CPU runs too slowly, the GPU has to wait for the results. Vera is NVIDIA's CPU designed specifically for this workload.

Vera features 88 NVIDIA custom Olympus CPU cores and a maximum memory bandwidth of 1.2TB/s. NVIDIA claims that in tasks such as Agent operation, reinforcement learning, and data processing, Vera can achieve a task completion speed up to 1.8 times faster than an x86 CPU.

NVIDIA's hardware allocation for the Agent is also becoming clearer: the Rubin GPU handles large-scale model computations, the Groq LPU accelerates Token generation, and the Vera CPU is responsible for tool invocation and code execution. SpaceXAI is even planning to send Vera Rubin into space, as the first-generation Starmind AI satellite will be based on the optimized version of the Vera Rubin NVL72.