Nvidia confirmed that the Groq 3 LPX inference system, born out of a $20 billion licensing deal with startup Groq, has entered full production and will be deployed before year-end. The system, which integrates 256 language processing units in a liquid-cooled design, delivers up to 35 times the throughput per megawatt compared to conventional setups. Nebius will be the first to deploy the solution via its Token Factory platform. The eight-month timeline from the deal signing on December 24, 2025, to full production represents one of the fastest in AI infrastructure history.
Source: Read the original article
