SANTA CLARA, California, December 24, 2025 – Nvidia has reached an agreement to license technology from AI inference startup Groq and hire key executives, including founder Jonathan Ross, in a transaction valued at approximately $20 billion, according to sources familiar with the matter.
The deal, described by Groq as a non-exclusive licensing arrangement for its inference technology, will see Ross, President Sunny Madra, and other senior leaders join Nvidia to scale the licensed IP. Groq will continue operating independently, with finance chief Simon Edwards stepping in as CEO, and its GroqCloud service remaining uninterrupted.
Groq’s Language Processing Units (LPUs) specialize in low-latency inference, claiming up to 10 times faster execution and one-tenth the energy use compared to traditional GPUs for running pre-trained AI models . The startup, founded in 2016 by former Google engineers who developed the Tensor Processing Unit, raised $750 million in September at a $6.9 billion valuation.
Nvidia CEO Jensen Huang stated the partnership would “extend our platform to serve a broad range of AI inference and real-time workloads.” Nvidia declined to comment on financial terms, while Groq emphasized continuity for customers.
The transaction represents Nvidia’s largest to date, surpassing its $6.9 billion Mellanox acquisition in 2019. It follows similar talent-and-IP deals, such as Nvidia’s $900 million arrangement with Enfabrica earlier this year.
Inference, the phase where AI models generate outputs, is projected to dominate future compute demands as deployment outpaces training. Groq’s deterministic architecture addresses this, positioning it as a complement or rival to Nvidia’s GPUs.
The deal awaits regulatory review, expected in early 2026. It underscores consolidation in the AI chip sector, where Nvidia holds a dominant share but faces challengers from startups and cloud providers developing custom silicon.
