DeepSeek is closing a funding round of roughly $7.4 billion — about 50 billion yuan — at a pre-money valuation of around $74 billion, according to reporting from Reuters, the South China Morning Post and CNBC citing people familiar with the process. The company has hired banks and is preparing a listing on Shanghai’s STAR Market, with a filing possible as early as late 2026 and a debut as early as the second quarter of 2027.
Tencent is reported to be weighing roughly 10 billion yuan and battery maker CATL about 5 billion yuan, which would make them the largest outside backers. NetEase, JD.com, IDG Capital, Monolith, Shixiang Capital, CPE, Legend Capital and a set of local-government vehicles are also named. DeepSeek has not confirmed any of it.
The size of the cheque is not the interesting part. The terms are.
Two classes of money
Per the reporting, the commercial investors — Tencent, JD.com and CATL among them — accepted a five-year lock-up and zero voting rights. China’s National Artificial Intelligence Industry Investment Fund, a state vehicle, invested directly, with voting rights and no lock-up.
Read that as a cap-table document rather than a headline. Some of the largest private cheques going into one of the world’s most valuable AI labs are buying pure economic exposure: no governance, no exit for half a decade. The state cheque buys governance and keeps its liquidity. That is not a rounding difference in deal terms. That is a statement about who the company answers to.
It is also a rational trade on both sides. DeepSeek gets capital without diluting control at exactly the moment an IPO would normally start forcing governance concessions. The commercial backers get a pre-IPO allocation in a name that is otherwise impossible to own. Nobody is being tricked. But anyone modelling DeepSeek’s future behaviour — on pricing, on open weights, on how it handles export and compute constraints — should weight the shareholder who can actually vote.
Our take: Ignore the valuation and read the governance. When commercial capital accepts no votes and a five-year lock while state capital takes votes and stays liquid, the round is telling you the company’s strategy will be set by policy priorities, not by return-seeking shareholders. That is a fine thing to invest alongside and a terrible thing to assume away. The same structure is worth checking for in every Chinese AI raise from here.
Mind the valuation spread
Be careful with the number itself. Reports over the past six weeks have put DeepSeek anywhere from about $59 billion to about $74 billion, with one outlet floating roughly $71 billion as an eventual STAR Market target. These are sourced figures on a private round that has not been announced, and they have moved as the round has been re-cut. The $74 billion pre-money figure is the one the most recent cluster of reports converges on — it is the best available estimate, not a confirmed price.
The direction of travel is clearer than the decimal. DeepSeek raised earlier this year, is raising again, and is doing it explicitly to fund model research and compute capacity ahead of an onshore listing. That sequencing — capital, then compute, then IPO — is the same one the US labs are running, executed on a domestic exchange and a domestic investor base.
What to watch
- Confirmation and the final number. The round was targeted to close by the end of August. A confirmed close, at a stated valuation, replaces six weeks of ranges.
- A STAR Market filing before year-end. That is the milestone that turns this from a fundraising story into a public-markets one, and it forces disclosure of the share structure.
- Whether the zero-vote structure spreads. Moonshot, Zhipu and MiniMax are all raising into the same IPO window. If their commercial investors take the same terms, this is a template, not a one-off.
- Tencent and CATL confirming their allocations. Both are listed companies. Commitments of this size tend to surface in filings.