Nvidia's Groq 3 LPX Chip Enters Full Production

Nvidia has launched its Groq 3 LPX inference accelerator into full production, unveiled at Hot Chips 2026. The chip is a purpose-built addition to Nvidia's Vera Rubin data center platform, designed to deliver ultra-fast token generation for agentic AI workloads. Neocloud provider Nebius Group is the first committed customer. In benchmarks by Artificial Analysis, Groq 3 LPX achieved a record 3,400 tokens per second running the Gemma 4 31B model, making it four times more responsive than rival platforms. The chip uses technology licensed from Groq Inc., acquired for $20 billion in December. SpaceX was also announced as a new Vera Rubin platform customer.