Nvidia Groq 3 LPX Enters Full Mass Production With Racks Online This Year, Boosting AI Inference Performance

TradingKey - Nvidia (NVDA) announced on Monday that the NVIDIA Groq 3 LPX, built on Groq technology, has entered full volume production, with related rack systems set to go live this year. Serving as a crucial component of the Vera Rubin platform, the Groq 3 LPX is mainly targeted at the rapidly growing AI inference market, further refining Nvidia's computing layout from AI model training to real-world deployment.

According to information released by Nvidia, each Groq 3 LPX rack contains 256 Groq 3 LPUs. Nvidia stated that in certain large AI model inference tasks, combining Groq 3 LPX with Vera Rubin can increase inference throughput per megawatt by up to approximately 35 times, while boosting the workload that AI data centers can process under the same power conditions.

At the end of 2025, Nvidia signed a non-exclusive technology licensing agreement with Groq and brought onboard Groq founder Jonathan Ross along with several core team members. Groq itself remains independently operated and continues to develop its GroqCloud business. In August this year, Groq officially became an Nvidia Cloud Partner.

Meanwhile, Groq itself is accelerating its adoption of Nvidia's new platform. The company announced on August 24 that it will be among the first cloud service providers to bring NVIDIA Groq 3 LPX and Vera Rubin NVL72 to market. Combined with Nvidia's confirmation that Groq 3 LPX has entered full volume production, this indicates that the cooperation between the two sides has gradually moved from technology licensing to product implementation and commercial deployment.

For Nvidia, the significance of Groq 3 LPX going live this year lies in further strengthening its competitiveness in the AI inference market. In the past, AI computing demand was mainly driven by large model training; however, as the user base of ChatGPT, Claude, and various AI agents expands, the inference demand generated by models running daily is growing rapidly. Nvidia is therefore expanding Vera Rubin from a pure GPU platform into a complete AI infrastructure that encompasses GPUs, CPUs, Groq LPUs, networking, and storage.

As Groq 3 LPX enters full volume production and is planned to launch this year, the partnership between Nvidia and Groq officially enters the commercialization phase. If the focus of computing growth in the next stage of the AI industry shifts further from "training models" to "running models", Groq's low-latency inference technology will serve as an important supplement for Nvidia to expand its AI chip market coverage.