NVIDIA Groq 3 LPX enters full production; Nebius first AI cloud to deploy

NVIDIA said its Groq 3 LPX inference accelerator is in full production as an extension of the Vera Rubin platform, aimed at faster token generation for agents. Nebius will add it to Vera Rubin NVL72 systems in its Token Factory service. The benchmark figures in the release, including 3,400 output tokens per second, are NVIDIA's own.

• Full production announced August 24, 2026
• Nebius is the first AI cloud to adopt it, in Token Factory alongside Vera Rubin NVL72
• 3,400 output tokens per second on Gemma 4 31B (NVIDIA-cited benchmark)
• Rack holds 256 LPUs