By Thorsten Meyer | October 30, 2026 | ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
For almost three years, DeepSeek was the AI lab that did not need anyone’s money. Its owner, the Chinese hedge fund High-Flyer, paid the bills, and its models kept up with American rivals at a fraction of the cost. That chapter is over. In June, DeepSeek closed its first outside funding round. Last week, several news organizations reported a second one: about 100 billion yuan, roughly $15 to $16 billion, one of the largest private rounds in China’s history.
Overnight, The Information added the detail that turns a big number into a bigger question. The round, it reports, brought in “a who’s who of Chinese tech investors—and some of the country’s most powerful state-backed institutions.”
I went through the reporting, the U.S. government order against DeepSeek from 20 October, DeepSeek’s privacy policy and its model licence. Here is what is confirmed, what is only reported, where reports disagree, and what it means if you or your company use DeepSeek.
Key takeaways
- Reported by several outlets, not confirmed by DeepSeek. Bloomberg, Reuters and The Wall Street Journal reported on 22 October that DeepSeek raised about 100 billion yuan (about $15 billion). The Information says 106 billion yuan ($15.9 billion) and that the round has closed. DeepSeek has not commented publicly.
- The price: about $74 billion before the new money. Reuters and the Journal put the pre-money valuation at about 500 billion yuan. For the June round, investor filings implied about 351 billion yuan, roughly $52 billion, after the money came in.
- Who leads is disputed. Bloomberg and Reuters name battery maker CATL and Tencent among the biggest investors. Newsletter summaries of The Information’s paywalled report say China’s state chip fund, the “Big Fund”, led the round with 27% state capital. I could not verify that, and no second outlet reports it.
- The state was already inside. After the June round, registry data reported by Caixin showed China’s National AI Industry Investment Fund as a direct shareholder with about 0.28%. Small, but it is a fund co-controlled by the state’s chip fund.
- The founder still controls the company, at least on the last reported register: about 91% of the registered capital sat with Liang Wenfeng, directly or through a partnership he controls. No registry change for the October round has been reported yet.
- The Tencent-and-Alibaba headline is six months old. The Information’s “$20 billion-plus” story is from 22 April, before the first round. Alibaba appears in no investor list for either round.
- For companies: the open model weights stay under the permissive MIT licence and can run on your own servers. DeepSeek’s own app and API store data in China, and U.S. federal civilian agencies must now remove DeepSeek products. Check which route you use. For users: nothing changes in the app today.
Products Worth Considering
Amazon Kindle Scribe Colorsoft 64GB (1st Generation)
DJI Mini 4K Fly More Combo
DJI Osmo Mobile 7P Phone Gimbal with Mic Mini Transmitter, Infinity Black
As an affiliate, we earn on qualifying purchases.
What happened, in order
DeepSeek is based in Hangzhou. Its founder is Liang Wenfeng, and The Information describes it as owned by the hedge fund High-Flyer. The fundraising took six months and two rounds, and the headlines are easy to mix up:
- 22 April: The Information reports that Tencent and Alibaba are in talks to invest at a valuation above $20 billion. DeepSeek had just started raising money for the first time. Days later, Yicai reports that Liang raised his direct stake from 1% to 34% with his own money.
- 16 June: The Information reports that the first round closed: more than 50 billion yuan ($7.4 billion) at a valuation above $50 billion, with an “unusual deal structure”.
- 17 July: Caixin, reading a stock-exchange filing by a small listed investor, puts the post-money valuation at more than 350 billion yuan (about $52 billion). Reuters and Caixin put Tencent’s share at about 10 billion yuan, CATL’s at 5 billion and Liang’s own money at 20 billion. By my reading of those figures, the founder was the biggest single investor in his own “outside” round.
- 25 July: Bloomberg reports that DeepSeek paused the second round after a purported transcript of Liang’s investor remarks went viral.
- 6 October: CNBC reports that DeepSeek may double the round to 100 billion yuan, with “state-backed funds and corporate investors jostling for stakes” and funds raised from private individuals turned away.
- 20 October: The U.S. Department of Homeland Security publishes an order removing DeepSeek products from federal civilian IT systems.
- 22 October: Bloomberg reports that DeepSeek raised about 100 billion yuan. Reuters, citing four people, says the round is “wrapping up”. The Wall Street Journal reports about $15 billion.
- 23 October: The Information reports the round has closed at 106 billion yuan, citing three people.
- 30 October: The Information publishes details of who is in the round, including “state-backed institutions”.
Infographic 01 · Two rounds in six months
From self-funded to $74 billion.
DeepSeek took no outside money until 2026. Then came two rounds, a paused deal and a U.S. order, all in six months.
How the price moved
Reported valuations, U.S. dollars
The striped bar is my calculation (about 600 billion yuan), not a reported figure. Exchange rates vary between outlets.
Who owned DeepSeek after round 1
Share of registered capital, July 2026, as reported by Caixin
Registered capital is not the same as economic value. The October round is not yet in any reported filing.
- 22 Apr 2026Tencent and Alibaba in talks, valuation above $20 billionThe Information · first round, talks only
- 16 Jun 2026Round 1 closes: more than 50bn yuan ($7.4bn), “unusual deal structure”The Information · Liang 20bn, Tencent ~10bn, CATL 5bn yuan (Reuters, Caixin)
- 25 Jul 2026Round 2 paused after a purported investor transcript went viralBloomberg
- 6 Oct 2026Round may double to 100bn yuan; state-backed funds and companies “jostling”CNBC
- 20 Oct 2026U.S. order: DeepSeek products out of federal civilian ITDHS, Federal Register 2026-21447
- 22 Oct 2026About 100bn yuan raised · Reuters: “wrapping up”, “not ironclad”Bloomberg, Reuters, WSJ
- 23 Oct 2026Closed at 106bn yuan ($15.9bn), three people sayThe Information
- 30 Oct 2026Investor details: tech titans and “state-backed institutions”The Information
Sources: The Information, 22 Apr 2026; The Information, 16 Jun 2026; The Information, 23 Oct 2026; The Information, 30 Oct 2026; Reuters, 22 Oct 2026; Bloomberg, 22 Oct 2026; WSJ, 22 Oct 2026; CNBC, 6 Oct 2026; Caixin, 17 Jul 2026; DHS order, Federal Register, 20 Oct 2026. Arithmetic marked as mine.
Powered by Thorsten Meyer AI
Where the reports agree, and where they don’t
They agree on the big picture: about 100 billion yuan, double the original target, at a pre-money valuation of about 500 billion yuan, with CATL and Tencent among the largest of more than 15 investors. Reuters names Midea, Meituan, Geely and the chipmaker GigaDevice, existing shareholders NetEase and IDG Capital, and the funds Boyu Capital, CPE and Legend Capital. The Wall Street Journal and The Information also report that Liang told investors in September that DeepSeek was earning revenue at an annualized pace of $1 billion. That is his own statement, not an audited figure.
They disagree on four points:
- How much. 100 billion yuan (Bloomberg, Reuters, the Journal) or 106 billion yuan (The Information). In dollars, that is $14.98 billion to $15.9 billion, partly depending on the exchange rate used.
- Done or nearly done. Reuters wrote that the round was “set to be complete in the coming days” and that “financial details and the investor lineup are not ironclad and could still change”. The Information says it has closed. DeepSeek has confirmed nothing, and no registry change has been reported.
- Who leads. Bloomberg and Reuters point to CATL and Tencent. The claim that the state’s Big Fund led, with 27% state capital, comes only from summaries of The Information’s paywalled article. Worth remembering: in May, the South China Morning Post reported that state-linked investors, including affiliates of Big Fund III, were backing the first round. In the end, the founder put in the most money and the state AI fund held about 0.28% of the registered capital.
- The valuation headlines. “$20 billion-plus” and “$74 billion” do not contradict each other. They are six months apart.
Infographic 02 · What we know
Confirmed, reported, disputed.
Big money, few documents. Here is how strong each claim about DeepSeek’s second round is.
On the record
- DHS order (20 Oct): DeepSeek products excluded from federal civilian procurement and removed from civilian agency IT, including third-party products “developed using DeepSeek implementations”
- Privacy policy: personal data, including prompts and chat history, is collected, processed and stored in China
- Model page: weights of DeepSeek-V4.1-Flash-1017 under the MIT licence
- No federal rule puts DeepSeek on the Entity List (my Federal Register search, 30 Oct)
Own sources, not confirmed
- About 100 billion yuan, about $15 billion (Bloomberg, Reuters, WSJ; The Information: 106bn)
- About 500 billion yuan, about $74 billion, before the new money (Reuters, WSJ)
- CATL and Tencent among the biggest of 15+ investors (Bloomberg, Reuters)
- State-backed funds among the bidders (CNBC) and in the round (The Information)
- Revenue at an annualized $1 billion, per Liang to investors (WSJ, The Information)
Treat with care
- Closed or still closing: The Information says closed; Reuters said “not ironclad”
- Big Fund as lead with 27% state capital: only in summaries of a paywalled report
- How much of the round is state money: not established
- U.S. official’s claim of chip smuggling via shell companies: an allegation (Reuters, June)
DeepSeek, CATL and Tencent have not commented. Every deal figure comes from people who spoke anonymously.Thorsten Meyer · as of 30 Oct 2026
Sources: DHS order, Federal Register, 20 Oct 2026; DeepSeek privacy policy, 10 Feb 2026; DeepSeek-V4.1-Flash-1017 model page; Reuters, 22 Oct 2026; Reuters, 16 Jun 2026; Bloomberg, 22 Oct 2026; WSJ, 22 Oct 2026; CNBC, 6 Oct 2026; The Information, 23 Oct 2026; The Information, 30 Oct 2026.
Powered by Thorsten Meyer AI
Why the money matters
Compute. DeepSeek became famous for training strong models cheaply. Money buys data centers, power, people and domestic chips. It cannot freely buy the most advanced American AI chips, which U.S. export rules restrict. The Wall Street Journal reports that DeepSeek still relies on Nvidia hardware to train its most demanding systems and is working with Huawei to move training onto Chinese-made chips. Bloomberg reports that DeepSeek has started spending on data centers and on open models for use at home and abroad.
The chip question also carries legal risk. In June, Reuters reported that a U.S. interagency committee had approved adding DeepSeek to the Commerce Department’s Entity List, a trade blacklist, but that the administration held off to avoid tension with Beijing. A senior State Department official had said DeepSeek supported China’s military and intelligence operations and tried to use shell companies in Southeast Asia to obtain advanced U.S. chips illegally. That is an allegation, not a finding. When I searched the Federal Register this morning, the DHS order was the only federal document naming DeepSeek. No Entity List rule names it.
State influence. “State money” here means two different things. One is direct state capital: the National AI Industry Investment Fund was set up in January 2025 with 60.06 billion yuan of registered capital, and the third phase of China’s chip fund, Big Fund III, is one of its controlling partners. The other is “industrial” money from large companies such as Tencent, CATL, Meituan and Geely, which are listed or private businesses, not state agencies. A percentage alone also says little. In June, the state AI fund was the only outside investor that appeared directly on DeepSeek’s register. By my reading of the register figures Caixin reported, all other outside money sat in a holding vehicle, Hangzhou Chengli, set up for the financing. How the October investors will hold their stakes is not public.
Some simple arithmetic puts the size in proportion. If the whole round is new shares, 100 billion yuan on top of 500 billion buys about one sixth of the company. That is my calculation, not a reported figure, and the real split may differ. Together, the two rounds add up to more than $22 billion this year. That is still small next to the more than $100 billion each that OpenAI and Anthropic have raised, a comparison Bloomberg draws.
The open-model playbook. DeepSeek publishes its models as open weights, which anyone can download and run. Its newest release, DeepSeek-V4.1-Flash-1017, states on its model page: “This repository and the model weights are licensed under the MIT License.” That is one of the most permissive licences in software. Nothing reported so far says the new investors want that changed.
My own reading is that two forces now pull in opposite directions. Open releases are DeepSeek’s calling card and its route into markets abroad, where Bloomberg says it wants to grow. But a company preparing a stock market listing, with a reported $1 billion revenue run rate and a long list of large shareholders, faces more pressure to make money from its best models. The licence on the next major release is the thing to watch. Until then, the ecosystem impact is simple: a better-funded DeepSeek can train more and release more, and every release puts price pressure on everyone else, as the V4.1 Flash pricing showed this month.
What this means for companies
- Know which route you use. “Using DeepSeek” can mean three different things. DeepSeek’s own app or API: its privacy policy says it will “directly collect, process and store your Personal Data in People’s Republic of China” and may share data to comply with “government requests”. Open weights on your own servers: the data stays where you run the model, but control its outbound network traffic anyway, as I explained in the Qwen file-upload case. Through a cloud provider: that provider’s terms and data region apply.
- Check your U.S. federal exposure. The DHS order bars DeepSeek products from federal civilian procurement. It also orders the removal of DeepSeek products, and of “third-party products and services developed using DeepSeek implementations”, from civilian agency IT. The notice does not define that phrase. If you sell software to U.S. civilian agencies, find out where DeepSeek models sit in your product, and ask your lawyers.
- Keep sanctions screening current. DeepSeek is not on the Entity List today, but Reuters reported that a listing was approved in committee and put on hold. Add DeepSeek and its affiliates to the names your screening tool watches, and re-check when lists change. This is not legal advice.
- Update your vendor file. The owner list will change: new strategic shareholders, possibly state-linked ones, and a possible listing on Shanghai’s STAR Market, for which DeepSeek hired the state-owned bank CITIC Securities, Reuters reported in September. Reports point to 2027. Record the owner, data location, licence and any government orders for every model you use. The MIT terms of versions you already downloaded are published; future releases could come with different terms.
- Don’t over-read the money. The funding does not change a model’s quality or price today. Test models on your own tasks before you switch, and keep the allegations around how some Chinese models were trained in view, covered in the distillation report.
Infographic 03 · Using DeepSeek at work
Three routes, three different risks.
“We use DeepSeek” can mean three different things. The funding round changes the owner list, not the route your data takes.
1 · DeepSeek’s app or API
Your data goes to DeepSeek
- Data collected, processed and stored in China, per the privacy policy
- May be shared to comply with “government requests”
- Covered by the DHS order for U.S. federal civilian agencies
- Keep sensitive and customer data out
2 · Open weights, your servers
The data stays where you run it
- MIT licence on the published weights
- Block outbound traffic you did not approve
- Record the version you run; future releases may carry other terms
- U.S. federal work: the order also names third-party products built with DeepSeek
3 · Through a cloud or model host
The host’s terms apply
- Check data region and retention in the host’s contract
- Ask which model version and licence the host serves
- Ask how the host handles government orders and sanctions lists
Five checks for your vendor file
- Who owns the model maker today, and who after the October round and a possible listing?
- Where does prompt and output data live, and under which law?
- Which licence covers the exact version you use?
- Which government orders apply to your customers, such as the DHS order?
- Is DeepSeek on your sanctions-screening watch list, so a new listing triggers an alert?
General guidance, not legal advice. Ask counsel about your own contracts and customers.
Sources: DeepSeek privacy policy, 10 Feb 2026; DeepSeek-V4.1-Flash-1017 model page; DHS order, Federal Register, 20 Oct 2026; Reuters, 16 Jun 2026.
Powered by Thorsten Meyer AI
What this means for users
- Nothing changes in the app today. The same rule applies as before: DeepSeek’s policy says it stores the personal data it collects in China, and that includes your prompts, uploaded files and chat history. Do not paste health, money, ID or work secrets into it, or into any chatbot you do not trust with them.
- Running a DeepSeek model yourself is different. A model running on your own computer does not send your chats to DeepSeek. An app built on top of it has its own privacy policy, so read that one.
- This is not an investment signal. DeepSeek is a private company. A Shanghai listing is reported as a plan, not a date. Nothing here is investment advice.
- Expect more models. If the money goes where Bloomberg says, into data centers and open models, you should see more releases and continued price pressure across AI tools.
Honest limits
- No company statement. Every deal figure comes from people who spoke anonymously. Reuters could not reach DeepSeek, CATL and Tencent, and they did not respond to Bloomberg.
- Paywalls. For The Information, I could read only the headlines and the opening lines it shows publicly. For Bloomberg and The Wall Street Journal, I relied on their headlines and on summaries by Quartz and Investing.com. I read Reuters’ report in syndicated copies.
- The 27% figure and the Big Fund lead are unverified. I report them because they circulate widely, not because I could confirm them.
- Registry data are second-hand. The ownership figures come from Caixin and Yicai reading Chinese registry and stock-exchange filings. Caixin’s article is paywalled; the shareholder percentages come from the summary it shows publicly. I did not access the registry myself.
- Exchange rates vary, so dollar figures differ slightly between outlets.
- My arithmetic is marked as such: the one-sixth share, the $22 billion total and the founder being the largest investor in the first round.
Verdict
DeepSeek is no longer the self-funded outlier. The size of the second round is well corroborated by several outlets with their own sources. The state’s role is real, through a direct fund stake in June and, by The Information’s account, more state-backed institutions now, but its size is not established. For companies, the questions are the same as last week: where does my data go, which licence applies, which government orders apply. They deserve a fresh look now, because DeepSeek’s list of owners is about to get longer.
Sources
- The Information: Behind DeepSeek’s $16 Billion Funding: Chinese Tech Titans and State Players, Jing Yang and Juro Osawa, 30 Oct 2026 (29 Oct U.S. time; paywalled, headline and opening read)
- The Information: DeepSeek Raises Nearly $16 billion in Second Funding Round, Juro Osawa, Jing Yang and Qianer Liu, 23 Oct 2026 (opening read), and DeepSeek Aims to Close $7.5 Billion Funding Round by End-October, 24 Sep 2026
- The Information: Tencent, Alibaba in Talks to Invest in DeepSeek at $20 Billion-Plus Valuation, 22 Apr 2026 (opening read), and DeepSeek Closes Record $7 Billion-Plus Funding with Unusual Deal Structure, 16 Jun 2026 (opening read)
- Reuters: DeepSeek wrapping up $15 billion fundraising to fuel AI race, sources say, Kane Wu, Yantoultra Ngui and Eduardo Baptista, 22 Oct 2026 (syndicated copy)
- Reuters: US holds off blacklisting China’s DeepSeek, more than 100 firms deemed security risks, sources say, Karen Freifeld, 16 Jun 2026 (syndicated copy)
- Bloomberg: DeepSeek Raises $15 Billion in Challenge to Leading US Labs, 22 Oct 2026 (paywalled; read via Investing.com and Quartz), and DeepSeek said to tell backers of funding pause after viral posts, 25 Jul 2026 (via Fortune)
- The Wall Street Journal: Chinese AI Startup DeepSeek Raises $15 Billion Ahead of Potential IPO, 22 Oct 2026 (paywalled; read via Quartz)
- CNBC: DeepSeek considers doubling latest funding round to up to $15 billion, sources say, 6 Oct 2026
- Caixin Global: DeepSeek Reaches $52 Billion Valuation in Round Backed by Tencent, CATL, Guan Cong, 17 Jul 2026
- Yicai Global: DeepSeek Founder Raises Chinese AI Firm’s Registered Capital Amid Reported Talks on Alibaba, Tencent Investment, Liu Xiaojie, 28 Apr 2026
- South China Morning Post: DeepSeek to soon close first external fundraising in US$50b valuation: sources, 8 May 2026
- The Standard (Hong Kong): 60b yuan AI giant up and running, 21 Jan 2025
- U.S. Department of Homeland Security: Notice of Order Regarding Exclusion and Removal of DeepSeek Products From Federal Civilian Executive Branch Information Technology Systems and Procurements, Federal Register 2026-21447, 20 Oct 2026
- DeepSeek: Privacy Policy, last updated 10 Feb 2026, and DeepSeek-V4.1-Flash-1017 model page, retrieved 30 Oct 2026
- China Daily: DeepSeek nears $15 billion funding round backed by Tencent, CATL, 22 Oct 2026
Related on ThorstenMeyerAI.com: DeepSeek V4.1 Flash 1017: near-frontier agents at a tenth of the price · Did China’s newest AI models learn from GPT-6 Astra? · Cohere says Qwen tried to send a file to Beijing · IBM’s CEO says run AI yourself
