NVIDIAGroq 3 LPX has entered full production, and the Vera Rubin NVL72 system has been expanded, aiming to provide fast token generation for agentic AI systems. In the Gemma 4 31B benchmark, Groq 3 LPX achieved 3400 output tokens per second in 100,000-token long-context use cases, which is 4 times faster than the closest alternative platform. Nebius is the first AI cloud to adopt Groq 3 LPX, SpaceXAI will use NVIDIAVera CPU, and CoreWeave has also deployed Spectrum-X Multiplane.