NVIDIA Senior Director Dion Harris informed reporters that the Groq 3 LPX rack has entered full production and will be deployed later this year alongside NVIDIA's Vera CPU and Rubin GPU in neocloud Nebius. This rack marks the commercialization of Groq's asset technology, which NVIDIA acquired last December for $20 billion, its largest recorded transaction. Each liquid-cooled rack encapsulates 256 Groq chips, and NVIDIA states it can deliver 3,400 tokens per second, surpassing the 750 tokens of the "ultrafast" mode supported by OpenAI Cerebras.