China's AI firms tread treacherous path to profit
08/01/2026 12:00
The author is a Reuters Breakingviews columnist. The opinions expressed are her own. Updates to add graphic. By Robyn Mak HONG KONG, Jan 8 (Reuters Breakingviews) - Spare a thought for China's artificial intelligence upstarts. The industry may be booming, but large language model specialists like Zhipu and MiniMax 0100.HK are bleeding red ink, squeezed by high research and development costs, Darwinian price wars against deep-pocketed giants and immature business models. With venture capital funding hit by Sino-American tensions, both are tapping public markets unusually early. That piles hard to compute risks on investors. Zhipu, formally known as Knowledge Atlas Technology 2513.HK, debuted in Hong Kong on Thursday after raising $552 million at a $6.6 billion valuation. The six-year-old company focuses on offering large-model services for Chinese enterprises, a market expected to grow to 90 billion yuan, or $13 billion, by 2030, up from less than 9 billion yuan last year, per research cited in Zhipu's prospectus. Yet revenue lags far behind the sky-rocketing costs of training and launching cutting-edge models. In the first half of 2025, Zhipu's research and development bill surged 86% year-on-year, to $228 million - more than 8 times its top line. With an expected monthly cash burn of 327 million yuan, Zhipu had just over two years of runway before its funds ran dry. Its Hong Kong share sale stretches that to a little over three years. The problem is raising prices looks unfeasible given intensifying competition from giants Baidu, Alibaba 9988.HK, DeepSeek and others. Zhipu's stock opened up 3% on Thursday morning. The profit and loss statement is similarly grim at MiniMax, whose shares are set to debut in Hong Kong on Friday. The company, known for consumer apps, including an AI agent, video creator and audio generator, might eventually be able to compete with Western giants overseas in new segments like AI coding. But that's a long way off: revenue was just $53.4 million in the nine months to end-September, less than a third of its R&D expenses for the period. To compare, annualised 2025 revenue for OpenAI and Anthropic is expected to be more than $13 billion and $9 billion respectively. In the past, deeply unprofitable companies like MiniMax and Zhipu would have been considered too-early stage for public markets. However, raising private capital is increasingly difficult against a sharp decline in foreign direct investment into the People's Republic, especially from U.S. investors, once a key support for startups. Government funding, meanwhile, is being channeled toward "hard tech" like semiconductors and aerospace. For public investors, as for venture capitalists, the odds remain heavily stacked against them. Follow Robyn Mak on X. CONTEXT NEWS Chinese artificial intelligence firm Knowledge Atlas Technology, also known as Zhipu AI, opened 3.3% higher than its initial public offering price on its first-day trading in Hong Kong on January 8. The company raised HK$4.3 billion at a valuation of about $6.6 billion. Zhipu made a net loss of 2.4 billion yuan in the first-half of 2025 on 190 million yuan of revenue. Rival MiniMax is expected to raise HK$4.2 billion in its Hong Kong initial public offering, Reuters reported on January 5, citing sources. The company is set to price its shares, due to start trading on January 9, at the top of the marketing range, at HK$165 per share. That would value MiniMax at about $6.5 billion. MiniMax made a net loss of $512 million in the nine months to end-September 2025 on $53.4 million of revenue. China's AI startups are bleeding red ink https://www.reuters.com/graphics/BRV-BRV/lbvgmyjyovq/chart.png (Editing by Una Galani; Production by Aditya Srivastav) ((For previous columns by the author, Reuters customers can click on MAK/ robyn.mak@thomsonreuters.com))