Luchen Tech Closes Hundreds of Millions of RMB in A++ Round | BlueRun Ventures

Luchen Tech has raised several hundred million yuan in an A++ round. The Beijing Artificial Intelligence Industry Investment Fund, Stonehill Capital, Capstone Capital, and Lingfeng Capital participated in the investment.

BlueRun Ventures was Luchen Tech's angel-round investor.

1. A new DataBricks platform for the large model era (https://cloud.luchentech.com/), maximizing LLM post-training and inference efficiency.
2. A world-class video generation foundation model (https://video.luchentech.com/).

Luchen's large model training and inference system Colossal-AI and its video foundation model Open-Sora both rank #1 globally on GitHub. In the open-source domain of AI training/inference systems and video models, both hold #1 spots worldwide by key metrics, with a combined 60,000 stars on GitHub. Luchen has also made significant strides in inference system optimization — its open-source inference system matches the latency of top-tier technologies like vLLM and JAX while delivering 50% faster inference speeds. A Fortune 500 company has made a substantial purchase of Luchen's enterprise inference software and achieved 3x efficiency gains.

Luchen's paid customer base is growing rapidly. Among pure software customers, clients including four Fortune 500 companies have already contributed tens of millions of yuan in revenue. Luchen Cloud's customer count is expected to reach 10,000 by year-end. The company's revenue growth is strong, with confirmed collections projected to hit 150 million RMB in 2025 and 300 million RMB in 2026.

Luchen Tech Confirmed Collections

On why Luchen Tech is building a video generation foundation model, founder and chairman Professor Yang You said:

"Why do we believe video foundation models are more likely to have long-term scaling laws than LLMs? Because video foundation models are trained on data that is a true reflection of the objective world, and the ultimate creator of that data is the Creator itself — the model will surely discover its underlying patterns. LLM training data comes from the internet and books, and the quality of its creators varies wildly, with plenty of ambiguity and noise.

Of course, most importantly, video foundation models will likely demand the highest compute optimization of any domain in the future. The super-app for video foundation models will ultimately be a contest of compute efficiency, and that's our advantage."

Machenike raises several hundred million yuan in C1 round to accelerate mass production of intelligent driving solutions and global market expansion | BlueRun Ventures

They say AI founders who come to the Buming Entrepreneurship Camp end up Booming | Registration countdown

BlueRun Ventures wins multiple awards from 36Kr, including Top 20 Industrial Investment Funds in China's AI and embodied intelligence sectors