DeepSeek's $15B Oversubscribed Round: Same Price, More Shares, No Voting Rights
The round ties DeepSeek's model roadmap to Chinese industrial players in batteries, chips, and autos, which gives it a capital-and-hardware pipeline outside the U.S. AI ecosystem. It also sets up a test for STAR Market rules: whether a no-voting limited partnership can be restructured in time for an early-2027 IPO.
DeepSeek's second private round began with a 50 billion yuan target, then climbed to 80 billion and settled near 100 billion yuan (about $15 billion) within days. Bloomberg's October 22 reconstruction shows the increases came after roadshow demand exceeded expectations, not from renegotiation. The pre-money valuation never moved: roughly 500 billion yuan ($74.8 billion) before and after the upsizing, so each new investor is buying more company for the same per-share price. The round is scheduled to close before a planned STAR Market IPO in early 2027.
CATL has committed around 15 billion yuan, double its first-round stake, and Tencent has added to its 10 billion yuan contribution; Geely, Shixi Capital, SMIC Juyuan, CPE Yuanfeng, Legend Capital and others are also in. Industrial investors are being picked over financial LPs: individual investors and private funds have been turned away. DeepSeek's stated uses for the money are model training, compute infrastructure, and talent, and the company has said it will restructure after the round, which is the step that unpacks the no-voting limited partnership structure before a public listing.
Liang Wenfeng is putting 20 billion yuan of his own money into the same structure. The round is expected to leave new and existing shareholders with roughly half each, and the deal is framed as the last private placement at current pricing before a 2027 IPO. Public confirmation is thin; Chinese media still use "plans to raise" and "expected to close," so the final amounts remain unverified.
The schedule change is driven by DeepSeek's own compute need: eighteen months ago it denied having financing plans, and six months later it had closed a 51 billion yuan first round and was raising again.
Keeping the pre-money valuation fixed while doubling the raise makes this a pre-IPO discount trade rather than a valuation bet: new shareholders pay the same price for a larger position before the STAR Market listing sets a public number.
Industrial investors are being admitted for what they can plug into DeepSeek: CATL's battery lines need AI dispatch, Geely brings autonomous driving, SMIC Juyuan brings chip manufacturing, and Tencent wants its cloud and inference stack on a domestic model.
The limited partnership control structure is stricter than a dual-class share setup: LPs hold economic rights only, without observation rights, and STAR Market's equal-rights requirement means the shell has to be broken before a 2027 filing.
The quiet closing strategy is a direct response to the first round's leaks; with no press release and no signing ceremony, DeepSeek and Tencent are trying to keep the final numbers unverified until the deal formally closes.
Chinese large-model financing has moved from survival to IPO sequencing, but the loss-making status of most AI companies leaves the 2027 listing depending on whether compute spend survives review as capitalized fixed assets.