Nvidia Secures Groq Inference Technology in $20 Billion Deal
Nvidia entered a $20 billion licensing agreement with Groq to integrate low-latency inference technology and hire its leadership team while avoiding antitrust scrutiny.
In late December 2025, Nvidia Corp. entered into a strategic agreement with AI chip startup Groq to integrate its low-latency Language Processing Unit (LPU) technology into the Nvidia AI factory architecture. While reports from CNBC and other sources value the transaction at approximately $20 billion in cash, the companies officially characterized the arrangement as a non-exclusive licensing agreement and asset transfer rather than a formal acquisition. This structure is widely viewed as a strategy to avoid the intense antitrust and regulatory scrutiny that typically accompanies the acquisition of a rival.
As part of the deal, Nvidia executed an acquihire of Groq's top leadership and approximately 90% of its workforce. Groq founder Jonathan Ross joined Nvidia as chief software architect, and President Sunny Madra became VP of hardware. Groq remains an independent entity focused on its cloud business, GroqCloud, under the leadership of new CEO Simon Edwards.
CEO Jensen Huang described AI inference as the industrial revolution of the industry, noting that the move allows Nvidia to diversify its supply chain by utilizing static RAM (SRAM) and reduce dependence on scarce high-bandwidth memory. The deal also serves to neutralize competitors like Cerebras Systems and AMD, and addresses performance bottlenecks for major clients such as OpenAI, which had reportedly explored alternative suppliers due to GPU response times. Nvidia plans to unveil the resulting inference computing platform at its GTC developer conference in San Jose.
