DeepSeek Closes a $15.8 Billion Round as State Capital Rises
Key Takeaways
- DeepSeek second round closed on 30 October 2026, raising about 105.5 billion yuan — roughly $15.8 billion — according to The Information, as reported by multiple outlets. DeepSeek itself has not publicly confirmed the figures.
- The round more than doubled its original 50 billion yuan target, drew nearly 30 investors, and put the pre-money valuation at about 500 billion yuan (roughly $74 billion), with a post-money figure of about 615.5 billion yuan (roughly $86 billion) in some reports.
- State-owned capital accounted for about 27% of this round, up from about 2% in the June 2026 round — the single most important change in the cap table.
- Tencent and CATL remain the largest backers, each contributing 10 billion yuan in this round, according to reports; CATL and its chairman together provided 15 billion yuan across the two rounds in some accounts.
- The figures come from a paywalled report summarised by secondary outlets, so every number in this article is attributed as reported, not as company-confirmed, and the valuation figures differ by source because they use pre-money and post-money bases.
Introduction
DeepSeek’s second funding round of 2026 is now, on the reporting available, closed — and the headline is not only its size. At about 105.5 billion yuan, or roughly $15.8 billion, it is one of the largest AI funding rounds ever reported in China, more than double its original 50 billion yuan target. But the number that will shape how the round is remembered is 27%: the reported share of state-owned capital, up from about 2% in the company’s first external round in June 2026.
That shift changes the story of DeepSeek. In June, the Hangzhou lab raised about 50 billion yuan (about $7.4 billion) in a round dominated by private and technology investors. Four months later, the cap table looks materially more state-weighted, with China’s national AI investment fund, state telecom operators, state banks and insurers all participating. DeepSeek has not publicly confirmed the figures, and the original report in The Information sits behind a paywall, so this article treats every figure as reported and flags where sources differ. But the direction of travel — a flagship Chinese AI lab drawing much closer to the state — is consistent across the outlets that have summarised the report.
This article sets out what is reported, what is still unconfirmed, who the major investors are, and why the state-capital shift matters for DeepSeek’s customers, competitors and for anyone trying to understand how China is funding its AI race.
What is reported, and what is not confirmed
The core facts, as reported by The Information on 30 October 2026 and summarised by AI Weekly, Tech Insider, BigGo Finance and others, are as follows. DeepSeek closed a second round of about 105.5 billion yuan, close to $15.8 billion, from nearly 30 investors. The round more than doubled its original 50 billion yuan target. The pre-money valuation sits at about 500 billion yuan, or roughly $74 billion, flat from the summer round according to AI Weekly. Some reports give a post-money valuation of about 615.5 billion yuan, roughly $86 billion — the sum of the pre-money valuation and the new money, which is why the two figures differ. They are not contradictory; they measure the company at different moments in the transaction.
The combined 2026 total now exceeds $22 billion, counting the June round of about $7.4 billion and this round of about $15.8 billion. The Information describes the raise, in AI Weekly’s summary, as “one of the biggest-ever fundraising deals in China’s AI sector”, with a cap table that includes “some of the country’s most powerful state-backed institutions”.
Two cautions apply to all of this. First, DeepSeek has not publicly confirmed the figures, so every number below carries the outlet that reported it. Second, at least one summary notes that it read only the headline and opening of The Information’s piece and built its figures from secondary summaries. Treat the state-share comparison — 2% to 27% — as a reported estimate until the original text or DeepSeek itself confirms it. That caution does not make the reporting worthless; it means the article’s job is to attribute, not to certify.
One further reported figure deserves the same treatment. Some coverage puts DeepSeek’s annualised revenue run rate at $1 billion as of September 2026, more than double its level a few months earlier, citing remarks by chief executive Liang Wenfeng to investors. That figure, if accurate, would put DeepSeek’s revenue growth in a different league from most open-weight labs. But it is an investor-remarks figure reported second-hand, not an audited result, and this article does not treat it as confirmed revenue.
Who paid in, and how the cap table changed
The investor list, as reported, is a mix of technology giants, state institutions and private capital. Tencent and CATL remain the two largest backers, each contributing 10 billion yuan in this round, according to BigGo Finance’s summary. CATL alone roughly doubled its earlier contribution, adding another 10 billion yuan on top of a previous 5 billion yuan ticket, according to AI Weekly. In another account, CATL and its chairman Zeng Yuqun provided 15 billion yuan combined across the two rounds, while Tencent contributed 10 billion yuan across the two rounds — a reminder that “largest backer” depends on whether the count is for this round alone or cumulatively.
State participation is where the round differs most from June. The China National Artificial Intelligence Industry Investment Fund contributed 1 billion yuan, matching its first-round commitment. China Mobile, China Telecom, CICC, CITIC Securities and China Life also invested, with individual contributions between 2 billion and 8 billion yuan in BigGo’s summary, or between 1 billion and 3 billion yuan for a wider set of state-linked names — including SMIC, ChangXin Memory, Meituan, Geely, vivo and Midea — in other summaries. The ranges differ because outlets group investors differently; the direction does not.
On a cumulative two-round basis, state-owned investors account for about 19% of the total, technology companies about 42%, with venture capital and private equity behind them, according to BigGo’s breakdown. That cumulative figure is useful because it stops the 27% from being misread as DeepSeek being state-owned. It is not. Even after this round, technology companies remain the largest cumulative group, and DeepSeek’s governance structure — in which, with the exception of the national AI fund, investors were required in the first round to inject funds into an entity held by Liang Wenfeng, give up voting rights and commit to a five-year share lock-up unless the company lists earlier — is designed to keep control with the founder. Whether the same terms applied in the second round has not been disclosed in the summaries available, and this article does not assume they did.
Other participants named in coverage include chipmaker GigaDevice, Meituan, Geely, Midea, NetEase, Boyu Capital, CPE, Legend Capital and IDG Capital, with more than 15 investors named in AI Weekly’s account and nearly 30 in total in BigGo’s. The discrepancy is a counting difference — named versus total — not a contradiction, but it is another reason to treat the investor list as reported rather than definitive.
Why the state share matters
A jump from 2% to 27% in state capital is not a bookkeeping detail. It signals, in the reading of multiple commentaries on the round, a distinct policy direction: the Chinese government directly underwriting its leading domestic AI enterprises through state-backed financing. That matters in three practical ways.
First, it changes DeepSeek’s access to compute and to domestic supply chains. State-linked investors in this round include telecom operators, a national AI fund and companies in the semiconductor supply chain, including SMIC and ChangXin Memory. For a lab that has faced export controls on advanced chips, domestic partners who control infrastructure, fabrication and distribution are strategically valuable in a way that a purely financial investor is not. The round should therefore be read alongside China’s draft AI export controls, issued on 16 October 2026, which placed frontier models and computing clusters under control — a reminder that the same state that invests in DeepSeek also regulates the inputs it needs.
Second, it changes how competitors and customers outside China will read DeepSeek. An open-weight lab with a largely private cap table is one thing; the same lab with more than a quarter of its latest round from state institutions will face different questions about data governance, procurement eligibility and export-control compliance from governments and enterprises in the United States, Europe and elsewhere. Those questions are not accusations, and this article makes none. They are the predictable consequence of a cap table that now includes the state at scale, and DeepSeek’s future enterprise sales outside China will have to answer them.
Third, it changes the competitive map inside China. DeepSeek’s June round, at about $50 billion pre-money, already made it one of China’s most valuable private AI companies. A second round at the same pre-money valuation, but more than twice the size and with state capital at 27%, suggests that the company’s backers — private and state alike — are willing to fund it at scale without demanding a higher valuation for the privilege. That is either a sign of discipline, a sign that the June valuation was already generous, or both. The reporting does not settle that question, and neither does this article.
What this means for DeepSeek’s models and users
For developers and users, the funding round does not, by itself, change any model. DeepSeek’s V4.1 Flash, shipped seven weeks before the round closed, remains the open-weights model that undercuts most Western rivals on price, and the round’s proceeds have not been tied in reporting to a specific model, chip purchase or product launch. What the round does change is the resources available for the next cycle of training, inference capacity and hiring — and, potentially, the constraints on how those resources can be used.
The timing also matters. The raise lands four months after the June round and seven weeks after V4.1 Flash, in a week when Anthropic was also arranging $63 billion in chip financing and the major Western labs were aligning on a training opt-out signal. The global AI funding cycle is not slowing; it is, if anything, becoming more state-shaped on both sides of the Pacific, with different mechanisms — private chip-lease vehicles in the United States, state-backed equity in China — serving the same purpose: securing compute at a scale that balance sheets alone cannot fund.
For readers who want the background on DeepSeek’s models, our explainer DeepSeek AI: What It Is, How It Works, Models, Features, Pricing, and More sets out what the lab builds and how its open-weight approach differs from closed frontier labs. And our coverage of cheaper AI models from OpenAI and Anthropic explains why price, not just capability, is the competitive dimension DeepSeek has used most effectively so far.
What to watch next
Watch for DeepSeek’s own confirmation, or a filing that corroborates the reported figures, and for the full investor list with per-investor amounts. Watch for whether the second round carries the same voting-rights and lock-up terms as the first, because that determines how much control the new state capital actually buys. And watch for what the money funds: a new model, a chip-supply arrangement, or inference capacity would each tell a different story about DeepSeek’s next move.
For now, the fair summary is this: DeepSeek has reportedly closed a very large second round, more than doubling its target, at a flat pre-money valuation, with state capital rising from about 2% to about 27%. The size is impressive. The cap-table shift is the story. And until DeepSeek confirms the numbers, every figure in that summary should be read with the word “reported” attached.
