DeepSeek nears $500M annualized revenue, plans 2027 IPO

A Chinese AI lab that spent most of last year being described as somehow doing this on a shoestring turns out to have a real business behind it. The Information reports that DeepSeek's annualized revenue is approaching $500 million, the first concrete revenue figure publicly attached to the Hangzhou lab.

The number lands in the middle of an unusually fast fundraising sequence. DeepSeek has already closed a $7.4 billion round (about 50 billion yuan) at a roughly $50 billion post-money valuation, and is reportedly in talks for another round of around $1.5 billion at a $71 billion pre-money valuation, per Bloomberg. Alongside the raise, the company is preparing a mainland IPO with a filing targeted as soon as late 2026 and a Shanghai debut planned for 2027.

Why revenue specifically matters here is the case for that IPO. Onshore listings need a story that regulators and public investors can underwrite, and cheap model plus giant valuation is not that story. Half a billion in annualized revenue, if it holds up on closer inspection, is the number that lets bankers actually put a multiple on DeepSeek and stop hand-waving about strategic importance.

The honest caveats stack up quickly. This is a single-source revenue disclosure, not audited financials, so take the specifics as reported rather than settled. The corporate structure is unusual: founder Liang Wenfeng reportedly holds around 78% of the equity, and investors in the recent round put their money into a limited partnership managed by him with a five-year lock-up and no voting rights. What the reporting does not give you is gross margin, cash burn, the split between API, enterprise and consumer revenue, or how exposed the top line is to the inference pricing compression happening across the whole market.

The people to watch are Tencent, CATL and JD.com on the cap table and the CSRC in Beijing on the approval side. If DeepSeek does list in Shanghai in 2027, it becomes the scarce liquid Chinese AI equity, and every other domestic lab suddenly has to explain why it is not yet in the same conversation.