A major consolidation has occurred in the AI chip industry: on December 25–26, 2025, Groq officially announced an agreement with NVIDIA. Instead of a full acquisition as previously rumored, the parties agreed on a non-exclusive licensing of Groq's unique LPU (Language Processing Unit) architecture for accelerating inference. The most resonant part of the deal is the transfer of Groq's key leadership and lead engineers, including founder Jonathan Ross (creator of the TPU), to NVIDIA. Groq stated it will retain its autonomy and continue developing the GroqCloud service.
Experts call this the "acqui-hire" of the year. NVIDIA effectively neutralizes its main competitor in token processing speed by integrating its brain trust, yet avoids harsh antitrust scrutiny by keeping the Groq brand alive. For the market, this signals NVIDIA's intent to radically speed up its inference solutions by combining CUDA power with Ross's architectural breakthroughs. The fate of GroqCloud's independent business remains uncertain, but the company secures significant immediate capital.
Source: ReutersGroqNVIDIAJonathan RossInferenceSemiconductors
