DeepSeek IPO lays bare China's shallow capital pool
BREAKINGVIEWS-DeepSeek IPO lays bare China's shallow capital pool The author is a Reuters Breakingviews columnist. The opinions expressed are her own. Refiles to fix style of the company name in paragraph five.
By Una Galani
HONG KONG, July 16 (Reuters Breakingviews) - China's AI pioneers are having to deal with a discomforting reality. They're raising money at a brisk clip, but a shortage of private capital, coupled with the soaring cost of building frontier models, is pushing them towards going public at market capitalisations far below those of U.S. rivals such as OpenAI and Anthropic. For Beijing, that presents both a challenge and an opportunity.
DeepSeek is a case in point. Founder Liang Wenfeng is preparing a Shanghai IPO despite his preference for pursuing artificial general intelligence over monetisation. The three-year-old company, which is seeking a $74 billion valuation in a fresh private funding round, charges the least among frontier model providers and only introduced peak-hour surcharges on its V4-Pro API to enhance service stability.
So a listing would underscore the limited backing of existing investors: Liang funded it with profits from High Flyer, his quantitative hedge fund, but he is only worth $36 billion. Backer Tencent 0700.HK – a social messaging-to-gaming giant – uses DeepSeek's model but also has its own to finance, while further support from China’s state-run National Artificial Intelligence Industry Investment Fund is likely to be limited.
Other sources of money look shallow, too. Venture capital funding in the U.S. soared to $267 billion in the first quarter of 2026. Even after excluding OpenAI's $122 billion share, the amount would be a record, KPMG says. China's total sum was a paltry $20 billion.
Chinese households appear willing to plug the gap. Some 20% of turnover in Hong Kong's tech-heavy market is driven by mainland traders via the Stock Connect. Shares of Z.ai 2513.HK, or Zhipu, whose GLM-5.1 model dominates global open-source rankings, are up 1,200% since its January IPO in the city despite it bleeding cash. The $150 billion company sold another $4 billion of shares last week and is planning a Shanghai listing.
These mainland debuts might help Xi Jinping coax more Chinese wealth into stocks and facilitate financial returns to households hurting since the property market's crash. It could also spur some citizens to forgo circumventing capital controls to invest in U.S. AI names, as well as keep DeepSeek and its peers near the industry's frontier.
But while China's AI companies are worth less than their U.S. peers, their valuation multiples have further to fall if the tech bubble bursts. According to The Information, DeepSeek's annualised revenue recently hit as much as $500 million, equating to a 148 times sales multiple. At $900 billion, Anthropic's implies a multiple closer to 30 times. If the first big bets on China's stock market by the next generation of investors turn out to be duds, the country's capital pool may look even smaller.
Follow Una Galani on Linkedin and X.
CONTEXT NEWS
DeepSeek has begun preparing for an IPO and may file as soon as this year, Bloomberg reported on July 14, citing unnamed sources.
The startup is also seeking to raise more funds in the private market ahead of the debut, weeks after closing a $7 billion financing round, the report added.
DeepSeek is targeting a $74 billion valuation for its pending capital raise, The Information said on the same day, citing three unnamed people, adding that the company's annualised revenue reached between $400 million and $500 million recently thanks to sales of cloud-based access to its AI models.
(Editing by Hudson Lockett; Production by Ujjaini Dutta)
((For previous columns by the author, Reuters customers can click on GALANI/ una.galani@thomsonreuters.com))
Recent news on Z.AI
See all newsBeijing seeks tighter grip on AI exports amid US rivalry (updated)
Moonshot execs in talks to settle dispute with shareholder ahead of planned IPO, people say
China AI developers publish safety tests for just 3.6% of model releases, report finds
Chinese labs are wild cards in AI governance
Chinese labs are wild cards in AI governance