Groq: CEO History

Groq started as a chip company trying to beat GPUs at inference and became, by 2026, primarily an AI inference cloud. Its LPU put model weights in on-chip SRAM and let a compiler plan every step, which made GroqCloud fast enough to go viral in early 2024. That speed drew Nvidia, which in December 2025 licensed the technology for about $20 billion and hired founder Jonathan Ross, now Nvidia's chief software architect. The independent Groq that remains runs 13 data centers and serves millions of developers. The company is headquartered in Mountain View, California.

Groq makes money by selling AI inference. Most customers pay per million tokens to run open-weight models (such as Llama, Qwen, Kimi and Whisper) through GroqCloud, an OpenAI-compatible API with a free tier and paid developer and enterprise plans. Larger customers buy dedicated capacity or on-premises GroqRack clusters. In December 2025 Groq also received a one-time licensing payment from Nvidia, reported at about $20 billion, much of which was reported to flow to investors and departing staff. Since that deal Groq describes itself as a neocloud and has become an Nvidia Cloud Partner, so its data centers can host Nvidia hardware as well as LPUs. Groq does not publish revenue.