The company, which focuses on large language models for enterprise clients, has raised 2 billion yuan this year, from mostly government sources
Key Takeaways:
Zhipu AI is aiming to become the first of China’s “AI tigers” to go public, as its competitors pivot away from costly foundational model development
The company relies heavily on government support, but such a strategy also carries risks, as other AI leaders have learned
Recognition or general knowledge?
That’s the question confronting China’s bumper crop of AI startups these days, as they try to chart a path that keeps them clear of controversy and overreliance on state support while also providing strong growth potential. While many of the earliest players focused on recognition AI that has found one of its biggest applications in surveillance, a newer generation of startups is finding bigger potential – and less controversy – in knowledge-based large language models (LLMs) like those from OpenAI and, more recently, DeepSeek.
Now, one of those, Zhipu AI, is aiming to become the first among a new group of pure LLM companies to go public, taking its first step towards an IPO by filing documentation with the Beijing Securities Regulatory Bureau last month, according to financial media Caixin. If the process goes smoothly, Zhipu could become the first of China’s six ” AI tigers” raised on knowledge-based LLMs to tap the public capital markets.
The company is widely regarded as one of the most likely survivors among these six players in China’s fiercely competitive AI landscape. Their space in LLMs has been recently disrupted by DeepSeek, which has gained significant attention following its recent release of more cost-effective models, including its reasoning-focused R1.
Despite its recent sudden leap to fame, DeepSeek isn’t considered one of the six AI tigers in LLMs. Among the other five – MiniMax, Moonshot, Baichuan, StepFun and 01.AI – two have already signaled a strategic shift away from foundational model development, of which LLMs are a subset. 01.AI, founded by AI pioneer and prominent venture capitalist and former Google China chief Lee Kai-Fu, has indicated it will cease investing in the development of foundational models, which demands substantial investment in AI chips and R&D.
Instead, these companies are pivoting toward developing applications based on foundational models developed by other companies like DeepSeek. Baichuan, founded by Wang Xiaochuan, former CEO of search engine Sougou, has similarly indicated a strategic pivot, focusing on healthcare applications rather than continuing investment in foundational models.
While the other three companies haven’t commented on their strategies lately, Zhipu AI stands out with its unequivocal commitment to foundational model development. The company has recently doubled down on this approach, launching a general-purpose AI agent in March, powered by its proprietary GLM model. It has also taken the additional step of open-sourcing its foundational models, similar to DeepSeek.
State backing
Among the foundational AI tigers, Zhipu has demonstrated a noteworthy ability to raise funds. While its peers have been quiet on that front over the past year, Zhipu has emerged as an investor favorite, securing at least 2 billion yuan ($276 million) in fresh funds this year alone.
The fundraising streak began in early March when the municipal government of Hangzhou, capital of Zhejiang province, emerged as the lead investor in a 1 billion yuan funding round. This was followed by investments from three other Chinese cities: Zhuhai in Guangdong province contributed 500 million yuan; Chengdu, capital of Sichuan province, invested 300 million yuan; and most recently, the Beijing municipal government provided 200 million yuan.
The recent funding wave is noteworthy for being led by government-backed sources, as China moves aggressively to develop the sector. Zhipu’s seizing on that largess marks a shift for the company from its previous reliance on private capital.
Founded in 2019 by Tsinghua University alumni Tang Jie and Zhang Peng, Zhipu AI began within the university’s walls and was later spun off. Its early funding rounds attracted prominent private investors like HongShan – formerly Sequoia China – and Chinese internet giants such as Alibaba and Tencent. Its roots at China’s most famous sciences university may be at least partly behind the recent flood of state capital to the company.
Zhipu was among China’s pioneers in LLM development, founded before OpenAI’s transformative release of ChatGPT in late 2022. The ChatGPT launch sparked an AI rush in China, leading to the emergence of numerous companies racing to develop Chinese alternatives. All of Zhipu’s major competitors were established in this post-ChatGPT era.
The company’s financial prowess is evident in its fundraising history. According to research firm RimeData, Zhipu has completed 14 funding rounds since its inception, raising over 9 billion yuan. Its ability to raise so much money, combined with its early focus on LLMs, has given Zhipu a leg up on its competitors in China’s intensely competitive AI landscape.
However, Zhipu’s real competition isn’t coming from other AI startups, but rather from tech giants like Alibaba, ByteDance, and Tencent, as well as the fast-rising DeepSeek. While Zhipu boasts strong backing from both public and private sources, such established players command even greater financial resources from their highly profitable core businesses.
While AI commercialization remains in its early stages, the race to capture consumer market share reveals the current state of the market. Zhipu’s Qingyan consumer app ranks ninth with 9 million monthly active users, significantly behind market leaders. Alibaba’s Quark app leads the pack with 149 million monthly active users, followed by ByteDance’s Doubao with 107 million and DeepSeek’s 96 million, according to April data from AI product tracker Aicpb.com.
That said, consumer applications aren’t Zhipu AI’s primary focus. The company has positioned itself more strategically as a provider of customized AI solutions for enterprise clients and government organizations, which tend to be more profitable than consumers.
In China’s technology market, government organizations, along with national and local state-owned enterprises, are some of the biggest spenders on software and digital solutions – which could work to Zhipu’s advantage due to its strong government connections.
But such heavy government reliance carries its own risks. A cautionary tale can be found in the fate of the four “AI dragons” from the pre-ChatGPT era, when AI was primarily associated with recognition technology, in this case involving computer vision.
The four AI dragons – SenseTime (0020.HK), Megvii, CloudWalk (688327.SH) and Yitu –primarily built their businesses on government work, providing surveillance systems powered by computer vision technology. However, these companies have largely faded from prominence following government spending cuts in surveillance-related projects.
Unable to diversify their customer base or expand into new applications, these once-prominent AI companies now find themselves struggling with dwindling business and mounting losses. In a nod to that reality, SenseTime has recently been rapidly shifting its business away from its recognition technology roots, and moving towards foundational models with better commercial potential.
At the end of the day, Zhipu’s substantial financial resources position it as a likely survivor as it competes against tech giants in China’s fast-evolving AI landscape. Still, it faces a critical challenge as it prepares to become first in its class to list: demonstrating to investors it can generate sustainable revenue from its proprietary products and services.