DeepSeek has begun preparing for an initial public offering in mainland China and may file as soon as this year, Bloomberg News reported, citing people familiar with the matter. The Hangzhou-based AI lab has started working with accounting and banking advisers to prepare its financial statements.
Ahead of that, it’s planning a new private fundraising round at a valuation of about 500 billion yuan, roughly $74 billion, Reuters reported, citing people with knowledge of the matter, just weeks after closing its first-ever outside round, which raised about $7.4 billion at a post-money valuation of roughly 450 billion yuan.
BREAKING: DeepSeek is preparing for an IPO and may file as soon as this year, per Bloomberg.
— The Kobeissi Letter (@KobeissiLetter) July 14, 2026
Details include:
1. The company is reportedly in talks with accounting and banking advisors and the IPO is expected to be listed in China
2. DeepSeek has begun talks for a fresh…
In that round, only China’s state AI fund got real ownership. Tencent, battery maker CATL, NetEase, and JD.com were among the other participants, but their capital went into a limited partnership Liang Wenfeng controls, carrying no voting power and a five-year lock-up. Liang personally committed 20 billion yuan to the round, which was also reportedly oversubscribed.
The valuation surge has made Liang, who still holds roughly 78% of the company, the wealthiest AI founder in the world on paper, with a stake worth about $36 billion, according to the Bloomberg Billionaires Index, ahead of Anthropic‘s Dario Amodei and OpenAI‘s Greg Brockman.
DeepSeek’s annualized revenue has climbed to between $400 million and $500 million recently, driven by sales of cloud access to its models, a factor that people close to the fundraising said is helping its case with investors. For comparison, OpenAI was last valued at $300 billion and Anthropic at $200 billion in their most recent rounds.
DeepSeek has built its position on open-source reasoning models that perform close to top US labs despite export controls limiting China’s access to advanced Nvidia chips; its cloud service instead runs on Huawei hardware, and the company has quietly ramped up hiring of chip-design engineers to develop its own AI inference chip.
In June, the company accounted for nearly 23% of all tokens processed by enterprise AI gateway Vercel, compared with Anthropic’s 32% share. Liang has told prospective investors the company will continue prioritizing long-term research and open-source releases over quick commercialization.