China Mobile and China Telecom Participate in DeepSeek's Second Round of 105.5 Billion Yuan Financing

November 3 news: "If you want to invest in DeepSeek, first submit a written statement explaining clearly what data support you can provide and what business synergies you can offer." This was the "test question" received by prospective investors in DeepSeek's second round of financing.

On October 30, The Information disclosed more details of this hundred-billion-level financing round: 105.5 billion yuan raised, a post-investment valuation of 615.5 billion yuan, nearly 30 investors, with China Mobile and China Telecom prominently on the list.

Why did the discerning Liang Wenfeng open the door to two telecom operators? The answer lies in DeepSeek's three criteria, and also in the operators' underlying strengths.

Hundred-billion financing, operators "take a seat at the table"

First, look at the fundamentals of this financing round.

According to The Information, DeepSeek's second financing round reached 105.5 billion yuan, with a post-investment valuation of 615.5 billion yuan, attracting nearly 30 investors. Tencent and CATL remain the two major lead investors: Tencent contributed 10 billion yuan in each round, totaling 20 billion yuan; the CATL system contributed 5 billion yuan in the first round and 10 billion yuan in this round, totaling 15 billion yuan. The lineup of new industrial capital entering this round is lavish—ChangXin Memory Technologies and its chairman Zhu Yiming, SMIC, vivo, Meituan, Midea, Luxshare Precision, Geely, Trip.com, Xiaohongshu, and others, with individual contributions mostly between 1 billion and 3 billion yuan.

The most noteworthy change is in state-owned capital. According to The Information's statistics, state-owned investors accounted for about 27% of this round's contributions, compared with only 2% in the first round. Across both rounds, the state-owned share reached about 19%. The National Artificial Intelligence Industry Investment Fund contributed 1 billion yuan in each round; China Mobile, China Telecom, CICC, CITIC Securities, China Life, and other institutions all appeared in this round, with individual contributions of 2 billion to 8 billion yuan. Among them, China Mobile plans to contribute 8 billion yuan, setting a record for an operator's investment in a domestic large-model company.

Where is the money going? The answer is almost written on the face of it. The combined funding of more than 150 billion yuan across the two rounds flows mainly into AI infrastructure: building large-scale intelligent computing clusters, purchasing high-end chips, and expanding inference service capacity. Liang Wenfeng said at the first-round investor meeting that DeepSeek's biggest future bottleneck is computing power. And with NVIDIA's high-end chip prices continuing to rise and concerns about an AI industry bubble bursting, the company has chosen to "stock up on grain while the window is open."

On the operating side, DeepSeek's confidence is also growing. According to media reports, Liang Wenfeng disclosed at an investor meeting at the end of September this year that the company's annualized revenue run rate had reached $1 billion. Regarding a listing, the company has hired CITIC Securities to prepare for an IPO on the STAR Market, and Yan Wentao, a former partner at Hillhouse Venture Capital, joined in September as chief financial officer.

Why does DeepSeek, which is not short of money, choose the operators?

For DeepSeek, money has never been a scarce commodity.

The company has been funded for years by its parent company High-Flyer Quant and is known for "no financing, no listing, no commercialization." When it completed its first financing round of about 51 billion yuan in June this year, Liang Wenfeng personally contributed 20 billion yuan, making him the largest single contributor; in this round he put in another 2 billion yuan. The enthusiasm of the secondary market also means DeepSeek is not short of ticket buyers—one investor once paid a 5 million yuan "meeting fee" and still could not meet Liang Wenfeng in person, and some returned empty-handed after three stakeouts at the Hangzhou headquarters. It is said that this round's investors were mainly selected personally by Liang Wenfeng, mostly by active invitation, and it is very difficult for those not invited to establish truly effective contact.

There are three criteria for becoming a DeepSeek investor: a mandatory five-year lock-up period, to filter out short-term money seeking quick profits; penetrating verification of the true contributors, to prevent competitors from entering through back channels; and priority for industrial synergy, with purely financial investors placed behind. In the first financing round, apart from the National Artificial Intelligence Industry Investment Fund, all external investors contributed through limited partnerships controlled by Liang Wenfeng, gave up voting rights, and accepted a five-year lock-up period. By the second round, the screening became further "exam-like"—prospective parties had to submit a written statement explaining what data support and business synergies they could provide to DeepSeek.

This may be why China Mobile and China Telecom were able to "take a seat at the table": not just "investing money," but more importantly "providing capabilities."

Computing power is the hardest card to beat. The stable, low-cost, large-scale computing power that large-model companies crave most is precisely the operators' family asset. China Mobile's total intelligent computing scale exceeds 107 EFLOPS, with the world's largest single intelligent computing center built by an operator, 12 regional intelligent computing centers, and 1,500 edge intelligent computing nodes; China Telecom's total "self-owned + accessed" intelligent computing scale had reached 118.8 EFLOPS by the first half of 2026, and it released the world's first commercial intelligent computing Ascend super node.

Delivery is the capability operators can see and touch. The most undervalued value of operators is the engineering capability to "deliver models to customers' machine rooms." In February 2025, the three major operators fully integrated DeepSeek; Mobile Cloud provides full-version, full-size DeepSeek services in 13 intelligent computing centers across the country, reducing inference costs by more than 13%, and also helped PetroChina complete full-stack domestically produced private deployment of DeepSeek V3/R1; Tianyi Cloud's "Xirang" platform became the first operator-level cloud platform in China to achieve full-stack domestically produced inference deployment of the DeepSeek model, and created China's first agriculture-industry large model based on DeepSeek.

Conversely, what DeepSeek brings to the operators is a ticket into the "Token business." China Telecom took the lead in launching a nationwide Token package, with Xingchen TokenHub bringing together 142 large models; China Mobile released the MoMA platform, aggregating more than 300 models for intensive Token operations—top-tier models like DeepSeek are precisely the core shelf items for "attracting customers" in the operators' AI service systems.

Taking a stake in DeepSeek does not equal binding oneself to DeepSeek

It should be emphasized that the operators are only taking stakes, not placing bets.

First, look at the investment landscape. Just half a year ago, China Mobile appeared on the investor list of Moonshot AI (Kimi)'s $2 billion Series D financing—Kimi, with a post-investment valuation of $20 billion, is a direct competitor of DeepSeek. China Telecom's moves have been even more frequent: in February 2026 it led the investment in the on-device large-model company ModelBest, and joined hands with Moore Threads, MetaX, DapuStor, Unitree Robotics and others to build out AI chips, storage and embodied intelligence; its subsidiary China Telecom Artificial Intelligence Company completed its first round of capital increase in 2025, bringing in four strategic investors including the National Artificial Intelligence Industry Investment Fund, becoming the first AI unicorn among central state-owned enterprises.

Next, look at their own models. China Mobile has persisted in self-developing the Jiutian large model; Jiutian JT4.1, released in August 2026, ranked 4th globally and 1st domestically on Artificial Analysis's comprehensive capability ranking of non-reasoning models, and was the first large model among central state-owned enterprises to pass the national "dual filing"; China Telecom's Xingchen large-model system has achieved "all modalities, all sizes, all domestically produced" and was selected as one of the "Top Ten National Heavy Weapons of Central State-Owned Enterprises for 2025."

Most critically, both operators follow a "multi-model, neutral" approach in model services: China Mobile's MoMA platform works in deep cooperation with leading model vendors such as Volcano Engine, Alibaba Qwen, Zhipu, MiniMax and Kimi, with flagship models launched on "Day 0"; China Telecom's Xingchen TokenHub brings together 142 large models, with mainstream models such as GLM-5.2 and DeepSeek going online on "Day 0."

So on the operators' shelves, DeepSeek is "one of them," not "the only one."

Liang Wenfeng chose the operators because things beyond money—computing power, delivery and scenarios—can only be provided by the operators; the operators invested in DeepSeek because they want the Token business and want to secure their position in the ecosystem.

When the most discerning entrepreneur and the most prudent state-owned operators sit down at the same card table, competition in China's AI industry has already moved from the second half of "burning money to buy models" into "competing on infrastructure, competing on implementation capability, competing on ecosystem synergy." And this two-way rush between DeepSeek and the operators is only just beginning.