DeepSeek 完成約 15 billion dollars 融資

DeepSeek Closes a 15 Billion Dollar Round — China's Largest AI Raise Ever — With an IPO in Its Pocket

DeepSeek has completed a funding round of roughly 100 billion yuan, about 15 billion dollars, according to multiple primary-market investors confirmed by the Securities Times — the largest single-round raise in Chinese AI history. Tencent and CATL returned as the biggest backers, but the story that has the industry talking is not the size. It is the entry fee: DeepSeek required every prospective investor to state in writing exactly what data support and business synergy it would bring, on the thesis that model-building is now an engineering problem whose real moats are training data and compute. Investors who could not answer got smaller allocations; those with deep industrial resources, like the late-joining Hengdian Capital, got large ones. With CITIC Securities advising an IPO targeted for Shanghai's STAR Market as early as next year, DeepSeek is positioned to beat both OpenAI and Moonshot to public markets — and to drag the entire Chinese AI supply chain onto its cap table.

What actually happened
The round was oversubscribed almost from the start. The initial target was 50 billion yuan; it was raised to 80 billion, then to roughly 100 billion — nearly double the original goal — with negotiations running from July through October. Tencent and CATL, both returning from June's first round, were the largest investors, with IDG Capital, Lighthouse Capital and Tenxiang Capital following on. The new money is strikingly industrial rather than purely financial: Legend Capital backed by Lenovo, Shixi Capital tied to memory-maker GigaDevice, Geely's automotive industry fund, the SMIC-affiliated Juxin Juyuan, and later Hengdian Capital. The A-share market piled in indirectly — Perfect World committed 150 million yuan across two dedicated funds (Lishi Xingshen at 7.52 billion yuan and Shixiang Shangshi at 3.65 billion yuan, together raising 11.17 billion yuan), Century Huatong's subsidiary put 650 million yuan into Shixiang Shangshi, and computing-lessor Litong Electronics subscribed 95 million yuan to Lishi Xingshen on October 22.

The terms are the real news. As one investor told the Securities Times, the underlying architecture of large models is no secret — the core barriers are training datasets and compute — so DeepSeek made resource collaboration a condition of entry. Perfect World brings game content, Litong brings compute leasing, Geely and CATL bring industrial data and manufacturing scale. It is a cap table assembled like a supply-chain pact.

The IPO clock
DeepSeek is not raising for the sake of raising. The company appointed Yan Wentao — a former Hillhouse Ventures partner with stints at Tencent Investment and H Capital — as CFO in September, hired CITIC Securities as IPO advisor, and is pushing to list on the STAR Market as early as the beginning of next year. Liang Wenfeng told investors in September that annualized revenue at current run-rate is about 1 billion dollars. The June round valued the company at roughly 500 billion yuan post-money; this round was struck at a pre-round valuation near 74 billion dollars. If the timeline holds, DeepSeek could become the first of the big model labs — ahead of OpenAI, ahead of Moonshot — to give public-market investors a ticker, and the STAR Market would get its defining AI listing. The company has also been spending: participating in CXMT's IPO subscription in July, a 141 million yuan strategic placement in Unitree in August, and a lead investment in a robotics startup planned for October.

Why this is exciting
Fifteen billion dollars of dry powder aimed at model training, compute build-out and talent is the single largest private bet ever placed on Chinese frontier AI — and it comes with something venture money rarely includes: guaranteed demand signals. CATL, Geely, GigaDevice, SMIC's ecosystem and Lenovo are not writing passive checks; they are pre-negotiating the data and compute partnerships that turn model capability into industrial deployment. The flywheel is the point. An investor base that spans batteries, autos, memory, foundry and compute leasing gives DeepSeek a vertically integrated testbed that most labs can only rent. For the open-source ecosystem that DeepSeek's releases feed, a capitalized, IPO-bound anchor is a strategic asset: it makes the open-weights path the best-funded path in China, not the charity path.

Why it is frustrating — if you are worried about a bubble
The frustrating counterweight is the arithmetic of expectations. A company with roughly 1 billion dollars of annualized revenue just raised 15 billion at a valuation near 74 billion dollars pre-round — and did so while the Wall Street Journal notes investors are growing more cautious about whether massive AI capital commitments convert into sustainable profit and whether rising valuations are justified. The same week, OpenAI was reported closing its own round of up to 30 billion dollars led by UAE investors at a 1.25 trillion dollar valuation. When the two most-watched labs in the world are raising sums that rival the GDP of small countries, the honest question is no longer "can they build it" — it is "who, exactly, is going to buy all of it, at what margin, for how long." The industrial-capital structure DeepSeek chose is a partial answer — money that comes with data and customers attached de-risks the revenue side — but it also means the fate of a frontier lab is now welded to the capital-expenditure cycles of battery plants and car factories. Concentration risk did not disappear; it changed shape.

Sources: Securities Times / e-Company, Wall Street Journal, Sina Finance, IT Home, Analytics Insight, threerouter