# Groq A Mountain View, CA-based AI chip company that designs custom Language Processing Units (LPUs) optimized for fast inference rather than model training. Founded in 2016 by Jonathan Ross, who previously architected Google's Tensor Processing Unit (TPU). Groq's LPU architecture delivers dramatically faster token generation and lower latency than GPU-based inference, positioning the company as a direct challenge to [[NVIDIA]]'s dominance in AI compute. ## Technology Groq's LPU is a purpose-built inference accelerator that uses a deterministic, software-defined architecture — the compiler schedules every operation at compile time, eliminating the runtime overhead that GPUs incur. This produces predictable, low-latency performance especially suited to large language model inference. The company operates GroqCloud, a cloud inference platform offering developers access to LPU-powered model serving. ## Funding and NVIDIA Deal Groq raised $750M in September 2025 at a $6.9B post-money valuation, more than doubling its $2.8B valuation from a year earlier. Investors include BlackRock, Neuberger Berman, DTCP, [[Samsung]], D1, and Altimeter. In December 2025, [[NVIDIA]] announced a non-exclusive licensing agreement reportedly worth approximately $20B for Groq's LPU inference technology. Founder Jonathan Ross and President Sunny Madra transitioned to NVIDIA as part of the deal, which was structured to avoid antitrust scrutiny — approximately 85% of shareholder payments were made upfront, with the remainder distributed through mid- and late 2026. Groq continues to operate independently under CEO Simon Edwards, formerly the company's CFO. ## KB Relevance Groq represents one of the most significant efforts to break [[NVIDIA]]'s effective monopoly on [[Artificial Intelligence]] compute hardware. The NVIDIA licensing deal is itself a telling indicator: rather than compete against LPU technology, NVIDIA chose to acquire access to it — a validation of the inference-specific chip thesis and a sign that the AI hardware landscape may be more contestable than NVIDIA's market share suggests. ## See Also - [[NVIDIA]] - [[Samsung]] - [[Artificial Intelligence]]