Last year, the AI chip startup Groq licensed its self-developed LPU inference technology non-exclusively to NVIDIA for a high price of $2 billion, and some core team members also joined NVIDIA. After this deal, the future direction of Groq was in question. However, within half a year, Groq provided an answer: transforming into an AI inference cloud service provider and completing a new round of funding of $650 million (about 4.4 billion yuan).

Confidence comes from "uniqueness"

Groq's confidence comes from a unique advantage: a global engineering team with practical experience in large-scale LPU operations. The LPU (Language Processing Unit) is a specialized inference chip developed by Groq, which excels at processing large model inference tasks with extremely low latency and has won the favor of many developers with its astonishing generation speed.

Although the technology was licensed to NVIDIA, the people and experience remain. Groq believes that this "muscle memory" is its core competitive advantage compared to other cloud service providers. Currently, Groq operates 13 data centers across four regions: North America, Europe, the Middle East, and Asia-Pacific, serving more than 5 million developers and thousands of AI-native companies, with weekly token consumption reaching trillions.

Focusing on 2027, betting on 200MW of computing power

This funding will be mainly used to expand AI inference infrastructure. Groq plans to deploy the latest inference technology and NVIDIA LPX system, aiming to increase computing capacity to 200 megawatts by the end of 2027, supporting larger-scale inference operations.

From a chip design company to an AI inference cloud service provider, Groq's transformation is not just a simple business restructuring. In a market dominated by NVIDIA, licensing technology to the strongest competitor while rapidly expanding its own cloud service footprint through its platform, this strategy of "retreating to advance" may be becoming a new survival wisdom for AI startups.