Nvidia has entered a non-exclusive licensing agreement with AI chip startup Groq in a deal reportedly valued at approximately $20 billion. The agreement centers on Groq’s inference-focused LPU (Language Processing Unit) technology, marking one of Nvidia’s most significant strategic moves as competition in AI hardware accelerates.

Groq’s LPUs are designed specifically for ultra-fast AI inference, emphasizing deterministic performance, lower latency, and reduced energy consumption compared to traditional GPUs. The company claims its architecture can deliver order-of-magnitude speed improvements for inference workloads while consuming a fraction of the power typically required by GPU-based systems. As inference increasingly becomes the dominant cost driver in large-scale AI deployments, this capability is drawing growing attention from hyperscalers and enterprises.

As part of the agreement, Groq CEO Jonathan Ross and President Sunny Madra will join Nvidia to help integrate and scale the licensed technology, while Groq continues to operate independently under CFO Simon Edwards. Ross is a well-known figure in AI hardware circles, having previously played a key role in developing Google’s TPU chips before founding Groq in 2016.

The timing of the deal is notable. Just months ago, Groq raised $750 million at a valuation of roughly $6.9 billion from investors including BlackRock, Samsung, and Cisco. Nvidia’s licensing move—rather than an outright acquisition—allows it to absorb Groq’s expertise and technology while maintaining strategic flexibility.

Looking ahead, the agreement reflects Nvidia’s broader defensive and expansionary strategy. As cloud giants like Google and Amazon push deeper into custom silicon, Nvidia appears focused on stockpiling both talent and complementary architectures to preserve its leadership in AI infrastructure. The Groq partnership strengthens Nvidia’s position in inference—a segment expected to grow faster than training as AI systems move into widespread production use.