[Musk Bets on NVIDIA Even for CPUs: SpaceXAI Deploys Vera to Run Agents] Beating AI News Flash: SpaceXAI will deploy NVIDIA's Vera CPU to run next-generation AI Agents. It is not intended to replace GPUs for large model computations but is primarily responsible for tasks between model calls, such as tool orchestration, code execution, data processing, task scheduling, and simulation. SpaceXAI will also continue using Vera Rubin to expand Grok's AI infrastructure. The more complex the AI Agent, the more work the CPU handles. After a model generates a tool call, tasks like executing code and querying data are typically handed off to the CPU. If the CPU runs too slowly, the GPU has to wait for the results to return. Vera is a CPU specifically designed by NVIDIA for these tasks. Vera features 88 NVIDIA self-developed Olympus CPU cores and a memory bandwidth of up to 1.2TB/s. NVIDIA claims that in tasks such as Agent operations, reinforcement learning, and data processing, Vera's task completion speed can reach up to 1.8 times that of x86 CPUs. NVIDIA's hardware division for Agents is becoming increasingly clear: Rubin GPUs handle large model computations, Groq LPUs accelerate token generation, and Vera CPUs manage tool orchestration and code execution. SpaceXAI also plans to bring Vera Rubin into space, with the first-generation Starmind AI satellite based on an optimized version of Vera Rubin NVL72. [Original Link]