NVIDIA Groq 3 LPX Inference Chip Enters Full Production

NVIDIA announced on August 24, 2026, that its Groq 3 LPX inference accelerator has entered full production, marking a significant milestone in the company's push to dominate the infrastructure for next-generation agentic AI systems. The chip, designed as a dedicated extension to NVIDIA's flagship Vera Rubin data center platform, is engineered to deliver ultrafast token generation speeds critical for highly responsive AI agents.

The announcement, made at the Hot Chips 2026 conference, follows NVIDIA's $20 billion acquisition of chip startup Groq Inc. in December 2025, the largest deal in the company's history. The Groq 3 LPX is built on technology licensed from that acquisition and is fabricated by Samsung. It features 500 megabytes of SRAM directly on the die to avoid memory bandwidth constraints that slow other inference accelerators. A full rackscale deployment can harness up to 256 LP30 accelerators, linked by high-bandwidth interconnects to work in tandem with GPUs.

In benchmark tests performed by Artificial Analysis, the Groq 3 LPX delivered a record 3,400 output tokens per second when running the open-source Gemma 4 31B agentic model with a 100,000-token context window....