Nvidia has hired the leader of a promising AI chip startup, it was announced Wednesday, as the artificial intelligence giant expands its technology empire.
Chipmaker Groq said the top executive's departure is part of a non-exclusive license agreement with Nvidia for inference technology, as both companies seek to expand access to low-cost AI processing.
Under the agreement, Groq founder Jonathan Ross, president Sunny Madra, and other team members will join Nvidia to help develop and scale Groq's technology.
Nvidia's dominance of the AI training chip market makes it the world's largest company by market valuation, but it faces increasing competition in the inference space from specialized startups like Groq.
AI inference refers to the process of running pre-trained AI models to make predictions or generate responses, such as when ChatGPT answers a user's question or an image recognition system identifies objects in a photo.
In a short statement, the company said Groq would remain an independent company under new chief executive Simon Edwards.
The statement came on the heels of a CNBC report that Nvidia had fully acquired Groq for $20 billion, but sources told AFP that no sale had taken place.
This arrangement is similar to an “acquisition.” This is an increasingly common practice in Silicon Valley, where large technology companies poach key staff from small businesses, leaving behind only a few remaining members of the company.
The practice is primarily aimed at avoiding the scrutiny of competition regulators, who are wary of tech giants acquiring up-and-coming rivals.
A recent example is Microsoft's 2024 deal with AI startup Inflection AI. The deal saw co-founder Mustafa Suleyman and much of the team join Microsoft, but the company remained independent.
Google is making a similar move, bringing in teams from AI startups like Character.AI in 2024.
Meta's 2025 deal to invest $14.3 billion in Scale AI and hire CEO Alexander Wang to head a new “superintelligence” AI lab is considered one of its biggest acquisition deals to date.
