According to foreign media reports, DeepSeek's second round of financing has expanded to CNY 105.5 billion, with a post-money valuation of approximately CNY 615.5 billion, involving nearly 30 investors. The most striking aspect of this round is not Tencent and CATL continuing to double down, but rather two telecom operators appearing simultaneously on the investor list.
It is reported that China Mobile plans to invest CNY 8 billion, while China Telecom plans to invest between CNY 2 billion and CNY8 billion, setting a record for the scale of telecom operator investments in domestic large model companies. A rising AI star and two telecom operators grappling with slowing traditional businesses sat at the same table. What does this deal mean for both parties?
For telecom operators, DeepSeek is an "entry ticket to the Token business."
The traditional growth logic of the three major telecom operators has reached its limit. In 2025, revenue growth rates were 0.91% for China Mobile, 0.07% for China Telecom, and 0.68% for China Unicom, all falling below 1%; in the first three quarters of 2026, China Mobile's revenue declined by 1.8% year-on-year, China Telecom by 3.59%, and only China Unicom saw a slight increase of 0.6%. Net profits fell by 7.9%, 14.84%, and 33.6% respectively, with computing power and intelligent services being the only bright spots in their financial reports. As internet traffic increasingly struggles to convert into revenue, operators are targeting the key concept of the intelligent economy—Tokens.
Recently, operators have been gradually pushing strategic concepts of Token management toward market implementation. China Mobile has established a "Token Office" at the group level, while China Telecom defines its intelligent cloud system as a "Token Management System." China Mobile has developed its MoMA model service platform, aggregating over 300 mainstream large models; China Telecom's Xingchen TokenHub aggregates 142 large models and has developed and promoted over 420 industry agents and 102 industry Skills.
However, these operators' self-developed models struggle to provide sufficient support. China Telecom's "Xingchen" and China Mobile's "Jiutian" have been deployed in areas such as intelligent customer service, cloud computing, and government/enterprise markets, but have failed to make significant waves in the market. Adding third-party models can continuously complement their own AI capabilities. Investing in DeepSeek is a preemptive positioning move. But operators are betting on more than just DeepSeek. China Mobile previously invested in Moonshot AI, and China Telecom led the investment in Modelbest.
For DeepSeek, it values operators' computing infrastructure and delivery advantages.
In this round of investment, DeepSeek required prospective investors to specify the data support and business synergy resources they could provide. This means that beyond financial strength, the chips, computing power, and industrial application resources brought by investors have also become important considerations for DeepSeek. DeepSeek needs stable, low-cost, and scalable computing power, which is precisely the advantage of telecom operators.
Operators possess massive computing infrastructure. China Mobile's total intelligent computing capacity exceeds 107 EFLOPS, having built 12 regional intelligent computing centers, upgraded 1,500 edge intelligent computing nodes, and completed two ten-thousand-card-level AI training bases in Hohhot and Harbin. China Telecom's "self-owned + connected" intelligent computing capacity reaches 118.8 EFLOPS, supporting over 20 heterogeneous chips, with 16 algorithms achieving integrated scheduling of computing power, network, and models, ranking first domestically in scheduling scale.
Operators have robust delivery capabilities. Mobile Cloud provides full-version DeepSeek services across 13 intelligent computing centers. Tianyi Cloud's "Xi Rang" is the first carrier-grade platform in China to achieve full-stack localized inference deployment of DeepSeek, applied in scenarios such as GPU cloud hosts, research assistants, and AI cloud PCs. Meanwhile, operators' government/enterprise channels and integration capabilities open up space for DeepSeek to reach industry clients. Previously, China Mobile assisted PetroChina in completing a full-stack localized private deployment of DeepSeek, serving as a proven sample.
For DeepSeek, operators' possession of data centers, backbone networks, intelligent computing centers, localized government/enterprise channels, and compliance systems will act as accelerators for converting model capabilities into industrial penetration rates.
Of course, this "mutual pursuit" requires a calm perspective.
By investing in DeepSeek, operators have purchased an entry ticket to the Token business, but whether they can turn it into a successful business after entering is another matter. Currently, the main models called by some Token packages launched by operators remain third-party models like DeepSeek. Industry insiders frankly state that if Token management merely replaces the "per GB" metric of the traffic era with "per Token," while model capabilities remain in others' hands, operators will find it difficult to truly escape their fate as mere pipes.
Equity stakes can secure priority supply and collaborative convenience, but cannot guarantee control over model capabilities. Operators currently retain their self-developed models and continue to push forward deployments in specific industries and customized services. How to form a complementary relationship between the general capabilities of third-party models like DeepSeek and the industry adaptability of self-developed models will be the direction the three major operators need to explore next. Additionally, whether collaboration can generate billable revenue is a key focus.
As AI competition shifts from "spending money to buy models" to the second half of "competing on infrastructure, delivery capabilities, and ecosystem synergy," the real test for operators lies not in whom they invested in, but in whether, after investing, they can transform themselves from "computing power landlords" into irreplaceable value nodes in the intelligent economy.
Editor/Layout: Zhu Wenfeng
Proofreader: Mei Yaxin
Supervisor: Liu Qicheng