China’s Zhipu AI Secures $140 Million Investment From Shanghai State Funds Amid IPO Push
By Liu Peilin and Denise Jia
Chinese generative artificial intelligence unicorn Zhipu AI has secured 1 billion yuan ($140 million) in fresh funding from Shanghai’s state-backed investors, boosting its momentum ahead of its planned initial public offering (IPO).
The announcement coincided with the unveiling of new products aimed at strengthening the company’s position as a leading domestic AI player.
At an ecosystem conference held Wednesday in Shanghai, Zhipu revealed Pudong Venture Capital and Zhangjiang Group — both state-owned entities — had invested in the company. The latest capital injection is a major milestone for the Beijing-based startup as it prepares to go public.
According to a disclosure on April 15 by China’s securities regulator, Zhipu’s parent company, Beijing Zhipu Huazhang Technology Co. Ltd., filed for IPO counseling the day before, appointing China International Capital Corp. as its sponsor.
Earlier this year, Zhipu attracted a flurry of investments from state-owned funds across Beijing, Chengdu, Hangzhou and Zhuhai. In March alone, the company raised more than 1 billion yuan in strategic funding rounds from investors such as Hangzhou Industry Investment Group and Hangzhou Shangcheng District State-owned Capital Operation Group Co. Ltd. Zhuhai Huafa Group added 500 million yuan to support Zhipu’s Generalized Language Model (GLM) foundation model development. Chengdu High-Tech Zone followed with a 300-million-yuan investment in the company, aiming to create Sichuan’s first major AI model. In April, Beijing’s AI industry fund injected another 200 million yuan.
According to Caixin calculations, before the latest Shanghai investment, Zhipu had already raised more than 10 billion yuan, reaching a valuation exceeding 20 billion yuan ($2.79 billion). Company filings indicate that founder Tang Jie and Chairman Liu Debing controlled a combined 36.96% of voting rights.
Founded in June 2019 and incubated at Tsinghua University, Zhipu is led by a core team — including CEO Zhang Peng and founder Tang — drawn from Tsinghua’s computer science department, where Tang was a professor. Among China’s “Six AI Tigers”, Zhipu is the oldest and largest, employing around 2,000 staff.
In August 2022, Zhipu became China’s first startup to launch a homegrown pre-trained large language model. Its app, Zhipu Qingyan, now has 25 million users with annual recurring revenue exceeding 10 million yuan. Zhang previously told Caixin the company aims to become “China’s OpenAI,” with plans spanning enterprise and consumer markets.
Commercial deployment has accelerated, focusing on enterprise clients. By March 2024, Zhipu had demonstrated 10 corporate use cases, with partners including Deloitte China, Focus Media, Huatai Securities, Mengniu Dairy, SAIC Motor and Kingsoft Office, mainly involving smart customer service, document parsing and content generation.
Zhipu is also expanding into consumer electronics by embedding its AI agents into devices. Since May 2024, it has announced deals with manufacturers such as Xpeng Motors, Honor, Samsung, Asus, and chipmakers Qualcomm and Intel to integrate its large model technology into hardware.
In October 2024, Honor’s new MagicOS 9.0 smartphone system featured AI-powered automation for tasks such as ordering coffee, managing subscriptions, setting app permissions and price comparisons, all built on Zhipu’s foundational models.
On November 29, Zhipu unveiled GLM PC, an AI agent for personal computers, upgrading its earlier mobile AutoGLM with reinforcement learning algorithms to enable more complex, cross-device, cross-application operations. However, Zhang acknowledged that while the system could theoretically execute any human-designed application after training, real-world performance still relies on users giving precise instructions due to PC complexity.
At Wednesday’s event, Zhipu announced two major updates: the open-sourcing of its new general-purpose visual-language model, GLM-4.1V-Thinking, which it said pushes the performance ceiling of 10-billion-parameter multimodal models, and the launch of a MaaS (Model-as-a-Service) intelligent agent aggregation platform to drive industry-specific AI adoption.
Contact reporter Denise Jia (huijuanjia@caixin.com)
caixinglobal.com is the English-language online news portal of Chinese financial and business news media group Caixin. Global Neighbours is authorized to reprint this article.
Image: Suriyo – stock.adobe.com