Zhipu has raised 3 billion yuan and is positioned as a domestic alternative to OpenAI, attracting state-owned capital.
Last week, Zhipu’s roadshow PPT circulated on social networks. It completed a new round of financing of 3 billion yuan. The investors include multiple strategic investors and state-owned capital, with existing shareholder Legend Capital and others following the investment. Jiejue Xingchen also announced several hundred million dollars in Series B financing. The financing war for large language models has entered the billions range. Currently, the main financing targets are state-owned capital and industrial leaders.
Zhipu’s PPT can be divided into four parts: technology, products, commercialization, and capital map. The technology section emphasizes the importance of large language models. Zhipu is one of the earliest in China. Zhipu showcases the timeline of large language model technology iterations. Starting in 2018, it marks key nodes. In its narrative, peers are not regarded as competitors. Those worthy of mention are the large language models launched by OpenAI, Zhiyuan, Zhipu, and Baidu. Zhipu was established in June 2019 and originated from the transformation of technological achievements of Tsinghua University. Initially, it studied knowledge graphs. In 2020, it focused on large language model algorithm research. In 2021, it designed the GLM algorithm and released the open-source 10-billion-parameter large language model GLM – 10B. Its path of large language model evolution is clear, benchmarking against GPT and being adapted to training and inference frameworks of domestic chips, telling the story of a domestic alternative to OpenAI.
Zhipu’s complete large language model full-stack technology architecture and model system benchmarking against OpenAI includes six elements, emphasizing comprehensive technical and product capabilities. Leading large language model companies have different development routes, like telling differentiated stories on the “dark side of the moon.”
In terms of commercialization, Zhipu constructs a three-layer business architecture of models, applications, and computing power. It has rapidly expanded to over 10,000 business customers on the B-side, covering four categories: technology and internet, vertical industry leaders, large state-owned enterprises and central enterprises, and automobiles and mobile phones. It uses cases to explain how to solve industry pain points. Zhipu claims to comprehensively benchmark against OpenAI, but its route is not completely the same as OpenAI. OpenAI became popular through its C-side product ChatGPT. Zhipu’s focus is not here. Its C-side product, the Zhipu Qingyan App, has more than 25 million users and an annualized income of over 10 million.
Finally, Zhipu talks about the market pattern and capital map. It emphasizes that it has accurately grasped its own characteristics and caters to the appetite of state-owned capital. It discloses that the large language model’s revenue is 170 million yuan, still far from profitability. The financing timeline shows that before ChatGPT exploded at the end of 2022, Zhipu had completed the Series B1 round of financing, with top financial investors and those from the Tsinghua University system adding their stakes. After the Series B1 round, the financing of large language model enterprises has accelerated. Zhipu has accumulated 4 billion yuan in financing. The first close of the Series B6 round has a pre-investment valuation of 20 billion yuan. The valuations of “Dark Side of the Moon” and Baichuan Intelligence also reach 20 billion yuan. Industrial giants and top financial investors have entered the game, and Zhipu also has the endorsement of state-owned capital. In November, Visual China’s announcement disclosed Zhipu’s equity structure. Finally, Zhipu looks forward to AGI. It has a distinction from OpenAI in its thinking on the endgame of artificial intelligence, emphasizing that AI serves humanity and telling a story more suitable for state-owned capital investors.
