🧠 Food for thought

Implications, context, and why it matters.

China’s AI IPO (initial public offering) runs on unproven models as losses exceed revenue

• Zhipu remains unprofitable, with 2024 revenue of about $42 million 1 and losses near $280 million 2. Its subscription plan costs about one-seventh of Anthropic’s Claude chatbot 1 (Anthropic is an AI lab), which keeps margins thin.
• MiniMax’s Talkie (a conversational AI app) brings in $70 million a year with 11 million monthly users 3. That business-to-consumer (B2C) focus differs from Zhipu’s government and enterprise push 4. Investors will ask which model can cover R&D needed to compete globally since OpenAI’s 2024 operating costs hit $8.5 billion 2.
• Hong Kong’s specialist tech listing regime (a local exchange pathway that lets early-stage tech companies list) may accept pre-revenue companies 5. Both firms face tougher investor scrutiny and need profit plans beyond subsidized prices for SME customers 2.

Cloud and Machine Learning Operations (MLOps) providers eye compute expansion

• If IPOs proceed, both will expand inference and training capacity across Hong Kong and Southeast Asia. Zhipu has over 2.7 million paid application programming interface (API) customers 1, driving compute demand outside mainland China.
• Zhipu has Alibaba backing 1 and took funding from Hangzhou state investors 6 (city-backed funds). MiniMax counts Alibaba and Tencent as backers 3. Those ties give GPU makers, clouds, and MLOps tools entry points. Post-IPO budgets will focus on scaling inference to meet the 150-fold jump in daily token consumption Zhipu said 6 (tokens are units of text processed by AI models).