Alibaba-backed AI startup Zhipu targets IPO as soon as 2025

Chinese tech firms are racing to build low-cost AI products after DeepSeek released a capable model supposedly trained at a fraction of the cost of American peers, including OpenAI

    • Zhipu is one of just a handful of Chinese startups trying to take on OpenAI globally.
    • Zhipu is one of just a handful of Chinese startups trying to take on OpenAI globally. PHOTO: REUTERS
    Published Tue, Apr 15, 2025 · 01:32 PM

    [NEW YORK] Zhipu has enlisted China International Capital Corporation (CICC) to spearhead an initial public offering (IPO) as soon as this year, angling to become the first of several prominent ChatGPT-competitors to tap public markets.

    The six-year-old firm and CICC have begun to prepare for the debut and could formally apply to list by October, according to IPO documents filed with the country’s stock markets watchdog.

    Backed by Alibaba Group Holding and Tencent Holdings, Zhipu is one of just a handful of Chinese startups trying to take on OpenAI globally. Since the emergence of DeepSeek, Zhipu and rivals from Moonshot to Minimax have rolled out low-cost artificial intelligence (AI) services at a rapid clip, aiming to stake out a spot in the burgeoning arena and set standards for future advancements.

    Zhipu in March closed a one billion yuan (S$346 million) financing round led by local government-backed firms, underscoring Beijing’s growing interest in AI following DeepSeek’s ascent. That investment came after Zhipu got added to another trade-restrictions blacklist by the US Commerce Department in January, which threatened to impede access to critical technology. The Chinese company has denied alleged links to the Chinese military.

    Zhipu’s IPO timing remains unclear. Many dealmakers are sitting tight while investors digest global trade and market volatility.

    But many AI startups require capital to sustain a dizzying pace of investment and research. Chinese tech firms are racing to build low-cost AI products after DeepSeek released a capable model supposedly trained at a fraction of the cost of American peers, including OpenAI.

    AI agents, which in theory can perform tasks for users, are one of the hottest areas of exploration.

    Beijing-based Zhipu has unveiled its own AutoGLM for deep research. The AI agent outperformed OpenAI’s GPT-4o and Anthropic’s Claude 3.5 Sonnet when accessed via a web browser, Zhipu chief executive officer Zhang Peng said during a presentation at the time. On Tuesday, Zhipu released another open-source model under its GLM series that it says costs users a fraction of what DeepSeek’s R1 charges.

    Not all players may ride out the intensifying competition. Chinese entrepreneur Kai-Fu Lee expects just three AI models to survive an impending domestic shakeup. BLOOMBERG

    Share with us your feedback on BT's products and services