Tencent’s Q3 Capital Expenditure Rises to RMB 81.433 Billion as Domestic Computing Power Rapidly Shifts Toward System-Level Competition
Tencent’s Q3 capital expenditure reached RMB 81.433 billion, up 527% year over year, with related prepayments used for Hunyuan model upgrades, WorkBuddy inference demand, and the development of AI capabilities across its products. Drawing on restrictions on advanced AI chip exports and IDC market data, the article notes that domestic computing-power competition is shifting from catching up in individual accelerator performance to system-level integration that coordinates chips, interconnects, cooling, and scheduling software.
Why read this
AI score 68
Tencent’s computing-power investment, chip export restrictions, and changes in domestic vendors’ market shares are examined within a single framework, providing context for understanding the shift in domestic computing-power competition toward system integration.
Topics
Body · AI translation
Article Body
Tencent’s capital expenditure surged 527% in the third quarter. Meanwhile, Southeast Asian remote computing-power channels have encountered fresh obstacles, potentially further accelerating the localization process. Institutions believe domestic computing chips are shifting from “catching up at the single-card level” to “system integration,” with a clear trend toward independence and controllability and substantial earnings upside expected for related companies.
Tencent Steps Up Computing Infrastructure Investment Again in Q3
On November 18, Tencent Holdings released its third-quarter 2026 results. The company generated total revenue of RMB 207.336 billion in Q3, up 8% year on year, and gross profit of RMB 115.911 billion, up 7% year on year.
Notably, capital expenditure reached RMB 81.433 billion this quarter, soaring 527% year on year and rising 54% quarter on quarter, far exceeding the market estimate of RMB 55.4 billion. Due to the elevated capital expenditure, the company recorded free cash flow of negative RMB 69.1 billion.
However, Tencent specifically noted that its operating cash flow included substantial prepayments for computing-power procurement to support upgrades to the Hunyuan model, WorkBuddy inference demand, and the development of AI capabilities across its products and services. Excluding these prepayments, the company’s free cash flow would have turned positive at RMB 16.9 billion.
The financial statements also showed that the company’s long-term “prepayments, deposits and other assets” surged from RMB 24.540 billion at the beginning of the year to RMB 162.901 billion, mainly to secure supplies of computing capacity and chips.
Regarding the external environment, the U.S. Bureau of Industry and Security (BIS) issued new rules restricting exports of advanced AI chips and servers to countries including Malaysia, while also limiting data centers’ large-scale provision of cloud services to China and other locations. The ceiling on computing-power exports to China by an individual data center or affiliated data centers is 12 million total processing performance, equivalent to approximately 750 Nvidia H200 chips.
In addition, Nvidia interim CEO Kress confirmed during the company’s latest earnings call that it would stop selling H200 chips to China.
According to IDC data, Nvidia’s share of China’s data-center AI chip market had fallen to 34% in the first half of 2026, with shipments down 44% year on year. Huawei Ascend accounted for 28%, followed by Alibaba’s T-Head, Cambricon, Kunlunxin, and Hygon Information Technology.
Notably, when leading large-model developers including DeepSeek, Zhipu AI, and Hunyuan recently launched new-generation products, they achieved same-day compatibility with chips from multiple domestic suppliers, shortening hardware-software coordination cycles to a matter of days.
Institutions: Domestic Computing Power Enters a Golden Development Period
Domestic computing chips are computational chips independently researched and designed by Chinese companies for artificial-intelligence training and inference scenarios. They include GPUs, DCUs, ASICs, and other types, and constitute the core hardware foundation supporting large-model training and inference and the construction of intelligent computing centers.
Soochow Securities said domestic supernodes remain in the early stages of large-scale deployment. Future competition will center on system-level integration capabilities—namely, the efficiency of computing-power utilization when scaling from a single card to a full rack or even a cluster—rather than merely whether domestic cards are available. Coordinated optimization across chips, interconnects, co-packaged optics, advanced packaging, power supply, cooling, and scheduling software will determine clusters’ effective computing power, interconnection efficiency, and cost per Token.
Regarding the rationale for accelerated domestic substitution in computing power, Huaxi Securities noted that domestic substitution has continued to advance over the past several years, while U.S. export controls have restricted Nvidia’s sales of high-end chips to China. Domestic chips therefore face relatively limited overseas competition, and orders are comparatively secure.
The institution further noted that the driving force behind domestic computing power has now shifted from “catching up in single-card performance” to “system-level engineering capabilities plus supply-assurance capabilities.” As cloud providers accelerate capital expenditure and domestic demand for computing power remains on a high-growth trajectory, structural mismatches in computing resources will benefit leading vendors capable of system-level supply involving complete supernode systems, optical interconnects, and liquid cooling. The trend toward accelerated localization is clear.
China Merchants Securities believes domestic computing chips continue to see rising deployment in fields such as telecommunications operators and intelligent computing centers, while their share among major internet companies is also growing rapidly. Based on its confidence in the development of the domestic computing-chip industry and senior policymakers’ determination to achieve independence and controllability, it remains firmly optimistic about industry trends in domestic computing chips, upstream foundry manufacturing and packaging and testing, AI support chips, and other fields.
Earnings at Multiple Concept Stocks Expected to Multiply
East Money’s concept-sector data show that approximately 40 A-share companies are currently associated with the AI chip concept, with a combined market capitalization of about RMB 2.25 trillion. Cambricon, Hygon Information Technology, MetaX Integrated Circuits, Montage Technology, Moore Threads, and others each have valuations exceeding RMB 100 billion.
Since November, the AI chip concept has performed relatively well. Spin Info Technologies led with a surge of nearly 46%, while Yuntian Lifei rose 24%. Aibulu, Hygon Information Technology, and Rockchip Electronics each gained more than 10%, while Montage Technology and Hesheng New Materials both fell more than 10%.
From a fund-flow perspective, East Money Choice data show that five AI chip concept stocks have attracted margin investors this month. Rockchip Electronics received an additional RMB 135 million in leveraged funds, while Spin Info Technologies and Anlogic Infotech attracted RMB 8.5 million and RMB 7.65 million, respectively.
In terms of future growth potential, consensus forecasts from three or more institutions indicate that earnings at five domestic computing-power concept stocks are expected to at least double this year.
Among them, institutions expect net profit at SigmaStar Technology to surge nearly 5.6-fold year on year, while net profit at Fudan Microelectronics could grow by more than 445%. Full-year earnings at Goke Microelectronics, Cambricon, and MetaX Integrated Circuits are also expected to at least double.
Cambricon is a core domestic AI computing-chip manufacturer. In the first three quarters of this year, it recorded net profit attributable to shareholders of RMB 3.715 billion, up 131.49% year on year, with a gross margin of 56.53%. The company recently said that demand in the AI computing-power market remains robust, supporting steady business growth, and that its gross margin is expected to remain relatively stable at a high level.
Hygon Information Technology is a major domestic manufacturer of high-end processors. Its revenue reached RMB 5.856 billion in the third quarter of this year, setting a new quarterly record. Research and development of the company’s next-generation Hygon 5 CPU and Deep Computing No. 4 DCU are proceeding along the established roadmaps, with overall progress in line with expectations. Significant upgrades are expected in metrics such as computing density and interconnection bandwidth.
(Article source: East Money Research Center) Middle of Article Embedded Advertisement
AI Trends