San Francisco: AI chip giant Nvidia has announced a $20 billion deal with AI high-performance accelerator chip designer Groq to license its technology, Groq said in a blog post. Besides the ‘non-exclusive’ license to Groq’s technology, its founder, Jonathan Ross, as well as Groq President Sunny Madra and other members of its engineering team are also expected to join Nvidia.
Groq, not to be confused with Elon Musk’s AI assistant Grok, is a market leader in the trained AI inference models market, which involves AI models that have already been trained to respond to user requests. Though Nvidia remains the market leader in training AI models and the manufacture of AI chips, Nvidia faces far more competition in inference models from rivals like AMD and Cerebras Systems.
Even though Nvidia has been at the forefront of the AI wave- its hardware powers most of the AI datacenters in the world- other companies have been quick to develop large sections of AI infrastructure.
For Nvidia, this acquisition is more about access to technology and manpower than the company as a whole- the deal is ‘non-exclusive’, offering a window to other players as well.
Groq has been one of the hottest startups to invest in, with its valuation reaching $6.9 billion in September following its latest funding round. Groq uses a different approach for its inference memory, using an on-chip memory called SRAM to speed up interactions with chatbots and other AI models, reducing the need for high external bandwidth memory chips.
With a high demand and a low manufacturing capacity, Nvidia is facing intense competition in the chip market with Google, Amazon, Meta and OpenAI also developing their own chips. This deal with Groq is aimed at maintaining the company’s lead, even as Groq will continue to operate as a separate company under Simon Edwards as CEO.
This deal follows a new pattern in recent years, where big tech firms are paying large sums for promising startups, but refraining from formally acquiring the company.
