Nvidia Puts Groq 3 LPX Into Production as SpaceXAI Adopts Vera CPU

Nvidia’s Vera Rubin is a rack-scale computing platform designed for agentic AI, combining the Vera CPU, Rubin GPUs and Groq 3 LPX inference accelerators. Unlike conventional chatbot workloads, AI agents must coordinate tools, execute code, process data and run simulations between model calls. That shifts more of the bottleneck to CPU orchestration and low-latency token generation, making tightly integrated infrastructure critical as AI systems move from training into interactive, commercial deployments.

On Aug. 24, 2026, Nvidia said SpaceXAI would deploy Vera CPUs as it expands the infrastructure behind Grok toward gigawatts of computing capacity, with an optimized Vera Rubin NVL72 also planned for its first-generation Starmind AI satellite. Nvidia separately put Groq 3 LPX into full production. In a Gemma 4 31B test using a 100,000-token context, the system generated 3,400 output tokens per second, four times the nearest alternative. Nebius will be the first AI cloud provider to add it to its Token Factory inference platform.