American AI chip company (not to be confused with xAI's model Grok). Groq's semiconductors are specifically designed for inference workloads—the process of running AI models—which has made them unappealing for many large AI labs that value flexibility between training and running models. The chips are well suited to reducing the cost of producing tokens, the fundamental unit of AI use.
In 2025 Saudi Aramco's digital arm agreed to obtain $1.5bn of Groq's chips, a deal that fed into Humain's strategy of offering cheap AI inference from Saudi Arabia.
In December 2025 Nvidia paid $20bn to license Groq's technology and hire its engineers. Nvidia subsequently unveiled the Groq 3 LPX, an inference chip using Groq's knowhow.
Ignorance must certainly be bliss or there wouldn't be so many people so resolutely pursuing it.