Groq licenses its inference technology to NVIDIA in a non-exclusive deal

The company behind the deal

Groq is a chip startup founded in 2016 by Jonathan Ross, a former member of Google's TPU team, and it pioneered the LPU (language processing unit) concept for inference chips ^[3]. Inference is the step in which a trained model produces an answer for a new input, as opposed to training, where the model's parameters are adjusted. Hardware built for inference is judged less on how fast it can train a model than on how quickly and cheaply it can return results, which is why the cost and latency of inference became a competitive battleground once models were put into everyday use.

What was announced

On 24 December 2025 Groq announced that it had entered into a non-exclusive licensing agreement with NVIDIA covering Groq's inference technology ^[1][2]. Groq said the agreement reflected a shared focus on expanding access to high-performance, low-cost inference ^[1]. The word "non-exclusive" matters: the arrangement does not hand NVIDIA sole rights to the technology, and it is not the same thing as buying the company.

Who moves, who stays

As part of the agreement, Groq founder Jonathan Ross, president Sunny Madra and other members of the Groq team joined NVIDIA to help advance and scale the licensed technology ^[1][2]. Groq itself will continue to operate as an independent company, with Simon Edwards stepping in as CEO, and GroqCloud, its cloud service, will continue to operate without interruption ^[1][2]. The announcement therefore bundles three separate things: a licence over the technology, the arrival of a group of people at NVIDIA, and a change of leadership at the firm that remains independent.

What had been reported before

Earlier reports had claimed that NVIDIA agreed to pay about $20 billion in cash for Groq's assets, citing Alex Davis, CEO of the investor Disruptive ^[3]. NVIDIA denied an acquisition, saying that it had only obtained a non-exclusive licence to Groq's intellectual property and hired engineers from the team ^[3]. No financial terms were disclosed ^[3]. The size of any payment is therefore not established by the public statements of either company.

Why it matters

证券时报 noted that the deal followed a pattern of big technology companies absorbing startup talent through licensing, a structure that limits antitrust exposure compared with an outright acquisition ^[3]. The distinction is worth spelling out for anyone reading deal news: an acquisition transfers ownership of a company, while a licence transfers permission to use technology, and hiring is a separate transaction again. Because those pieces can be signed separately, one announcement can be reported both as a licensing agreement and as the hiring of a founding team, with the company that supplied the technology still standing.

For readers watching the inference chip market, the arrangement is a compact case study in how talent, intellectual property and products can travel along different paths at the same time: the engineers go to NVIDIA, the licence covers the technology, and Groq keeps its name, its cloud service and a new chief executive ^[1][2][3]. What the agreement does not settle publicly is the money involved, since no financial terms were disclosed ^[3].