The vLLM team officially announced their entrepreneurship: They raised $150 million in financing, and YOU Kaichao, a winner of the Tsinghua Special Award, became a co-founder.
The cornerstone of large model inference, vLLM, has now become a startup.
Early Friday morning Beijing time, news came that Inferact, an artificial intelligence startup founded by the creators of the open-source software vLLM, was officially established. It raised $150 million (approximately 1 billion yuan) in its seed round of financing, and the company's valuation reached $800 million.
This round of financing was led by venture capital firms Andreessen Horowitz (a16z) and Lightspeed. Sequoia Capital, Altimeter Capital, Redpoint Ventures, and ZhenFund also participated in the investment.
Although Inferact's $150 million angel round financing is less than the $1 billion of Ilya Sutskever's company SSI, it has exceeded the $115 million of Mistral AI. It is one of the largest seed round financings in history, marking a rapid increase in the industry's attention to AI inference infrastructure.
Inferact's mission is to develop vLLM into the world's leading AI inference engine and accelerate the development of AI by reducing inference costs and speeding up inference.
The company believes that the biggest challenge the AI industry will face in the future is not building new models, but how to run existing models with low cost and high reliability.
Undoubtedly, the core of Inferact is the open-source project vLLM, an open-source project launched in 2023, aiming to help enterprises efficiently run AI models on data center hardware.
vLLM was initially developed by the Sky Computing Lab at the University of California, Berkeley (UC Berkeley) and is now managed by the PyTorch Foundation. It has attracted more than 2,000 contributors from the entire AI industry and is the most popular open-source large model inference acceleration framework globally.
Today, vLLM's inference capabilities support technology companies such as Meta, Google, and Character.AI.
Simon Mo, the CEO of Inferact, is a doctoral student at Berkeley and one of the founding maintainers of vLLM. Mo said that the company was founded in November 2025 and was officially announced this week. He compared the origin of Inferact with some early software projects at Berkeley, which later developed into larger enterprises, such as Apache Spark and Ray.
While announcing the financing, Lightspeed also released an interview with Simon Mo. In it, Simon Mo talked about his concerns about the global shortage of AI computing power. "The AI clusters currently used for large model training will be completely used for inference within six months... Inference will gradually consume all computing capacity and exhaust all newly added capacity."
In the announcement, Inferact said that it positions itself at the intersection of models and hardware: when model manufacturers release new architectures, they will cooperate with vLLM to ensure first-day support; when hardware manufacturers develop new chips, they will integrate with vLLM; when large model teams conduct large-scale deployments, they will run vLLM, from cutting-edge laboratories to hyperscale data centers, and even startups serving millions of users.
Today, vLLM supports more than 500 model architectures, can run on more than 200 accelerators, and supports global-scale inference. This ecosystem built by more than 2,000 contributors is the foundation for the establishment of Inferact.
Inferact said that its primary task is to continue to support vLLM as an independent open-source project and share the improvement results with the community. They plan to further improve vLLM's performance, deepen support for emerging model architectures, and expand coverage of cutting-edge hardware. Inferact's second goal is to develop an independent commercial product to help enterprises run AI models more efficiently on different types of hardware.
It is worth noting that Kaichao You, a Ph.D. from Tsinghua University and a core contributor to the vLLM project, has become a co-founder of this company.
It is reported that Inferact's founding team includes Simon Mo, Woosuk Kwon, Kaichao You, Roger Wang, Joseph Gonzalez, Ion Stoica, and others.
Reference Links
https://inferact.ai/
https://www.bloomberg.com/news/articles/2026-01-22/andreessen-backed-inferact-raises-150-million-in-seed-round
This article is from the WeChat public account "Almost Human" (ID: almosthuman2014). The author is Zenan. It is published by 36Kr with authorization.