--- title: "The Companies That Supply DeepSeek Now Own a Piece of It" date: 2026-11-02 category: China AI Decoded site: NeuroAI canonical: https://neuroai.site/a/na-decode-deepseek-cap-table language: en --- # The Companies That Supply DeepSeek Now Own a Piece of It > The reported RMB 105.5 billion round put a memory maker, a foundry and two telecoms on DeepSeek's cap table. In a shortage, equity buys allocation. The most revealing names in DeepSeek's second financing are not the funds. They are the companies that sit on the supply chain it depends on: a memory maker and its chairman, a chip foundry, an assembler, two telecoms, a battery giant. When the people who sell you your bottlenecks become your shareholders, the round stops being a story about growth. It becomes a story about who gets served first. ## Key takeaways - The Information reported on 30 October 2026 that DeepSeek raised RMB 105.5 billion (≈US$14.9bn / HK$116bn) at a post-money valuation of RMB 615.5 billion (≈US$86.7bn / HK$677bn). - State-linked investors supplied 27% of that money, against 2% in DeepSeek's June 2026 first round, per the same reporting, which the company has not confirmed. - ChangXin Memory (长鑫存储) and its chairman Zhu Yiming joined the investor list, as did SMIC (中芯国际); neither amount was disclosed. - Tencent put in RMB 10 billion in each round and CATL's group RMB 5 billion then RMB 10 billion, according to Chinese financial press. - DeepSeek's annualised revenue was about US$1 billion in September 2026, by Epoch AI's estimate, the same figure its founder gave investors. ## What was reported, and what has not been On 30 October 2026 The Information reported the round at about RMB 105.5 billion, with nearly 30 investors and a post-money valuation of RMB 615.5 billion. Chinese financial outlets repeated the figures within days. DeepSeek has published no statement, none of the investors named has confirmed, and no filing exists. Treat the numbers accordingly: the most detailed account available, sourced to people familiar with the transaction, and not audited. The first round, in June 2026, was about RMB 51 billion (≈US$7.2bn / HK$56bn). What is not in dispute is the shape of the list. It is not a list of venture funds with one strategic investor attached. It is a bill of materials. ## The list is the interesting part - **Memory.** ChangXin Memory, the domestic DRAM producer, and its chairman Zhu Yiming. Memory has been the binding constraint on Chinese accelerator output all year. - **Foundry.** SMIC (中芯国际), the country's largest chip foundry and partly state-owned, whose advanced-node capacity is the other half of that constraint. - **Networks and sites.** China Mobile and China Telecom, each reported at RMB 2–8 billion. They own data-centre campuses, power contracts and the government sales channel. - **Assembly and devices.** Luxshare, vivo, Midea, Geely, Ctrip, Xiaohongshu, each reported at RMB 1–3 billion. These are the routes to handsets, appliances, cars and traffic. - **Money.** The National AI Industry Investment Fund at RMB 1 billion, unchanged from the first round, alongside CITIC Securities, China International Capital Corporation and China Life. In a shortage, cash does not clear the market; allocation does. Equity is one of the few instruments that survives a shortage, because it converts money into a claim on next year's output rather than this quarter's spot price. The Information also reported that prospective investors were asked to state in writing what data support and business cooperation they could offer. That condition is unusual in a financial round. It reads like procurement wearing the clothes of equity. ## The state moved from guest to owner The number that will be quoted longest is the shift in state-linked money: 27% of this round, against 2% of the first. Read it precisely. It is a share of the money contributed, not of the equity, and it says nothing about board control. The National AI Industry Investment Fund's own cheque is small and unchanged at RMB 1 billion. The change is in the company around it — telecoms, an investment bank, an insurer, a state foundry. Those are institutions that invest when a company has stopped being a bet and started being infrastructure. ## Why more money buys less than it looks People cited by The Information said the fundraising target was raised during the process, partly because accelerator prices had risen. That detail matters more than the headline size: a round that grew while input costs grew faster is a defensive raise, not an expansionary one. Two reported facts support that reading. Bloomberg reported in September 2026 that DeepSeek planned to move a large share of its inference onto Huawei accelerators at a data centre in Inner Mongolia. Reuters reported that the company was working with Huawei on software for the Ascend platform. The strategy is legible: open weights for developer share, domestic silicon for supply, and capital sized to absorb price increases. Against about US$1 billion of annualised revenue, the reported post-money valuation is roughly 87 times sales at the fixed rate this archive uses (¥7.1 = US$1). Chinese financial press has also reported preparation for a STAR Market (科创板) listing, with CITIC Securities as counsellor. No filing has been published. ## What this means for buyers outside China Three things, none of them comfortable. First, open weights and open capacity are different things. The weights remain free to download; the servers behind the API now belong to a coalition with its own priorities. Second, price signals get weaker. An integrated stack can cross-subsidise inference, so a low token price tells you less about the true cost of the compute behind it. Third, this is a template. If a model lab can trade equity for memory, wafers and grid connections, others will try the same, and the Chinese AI stack will keep consolidating into fewer, more vertically joined groups. ## Honest limitations The entire financing account rests on one outlet's reporting from unnamed sources; no filing, statement or confirmation exists, so the figures could be revised or denied. Revenue is an external estimate corroborated by what the founder told investors, not a disclosed number. The 27% is a share of money raised, not of equity, and cannot be used to infer control. ## What readers can do now - Watch for the STAR Market prospectus. It would give DeepSeek an audited cap table and audited revenue for the first time, and settle most of what is currently reported. - Do not compute ownership from the 27%. Ask instead which investors hold board seats, which is the only published route to control. - If you build on DeepSeek's API, add an allocation-risk clause to your supplier review this quarter: integrated owners have reason to serve their own ecosystem first when capacity is short. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-decode-deepseek-cap-table Free to quote with attribution and a link to the original.