Tencent’s Q3 Capital Expenditure Surges 527.8% Year on Year, WorkBuddy’s Daily Token Usage Exceeds 100 Trillion
Tencent’s capital expenditure reached RMB 81.433 billion in Q3 2026, up approximately 527.8% year on year, as AI-related R&D investment and computing capacity purchases continued to rise. During the same period, Tencent Cloud service revenue grew 35% year on year, but new AI products including Hy, Yuanbao, CodeBuddy, WorkBuddy, and Xiaowei reduced non-IFRS operating profit by approximately RMB 15.2 billion. As of early November, WorkBuddy had more than 10 million daily active users and averaged over 100 trillion Token calls per day.
Why read this
AI score 81
The earnings report presents prepaid computing capacity, cloud business growth, and the drag on profitability from new AI products side by side, offering a way to assess the costs and commercial returns of Tencent’s increased AI investment.
Topics
Body · AI translation
Capital spending soars 528%! Tencent doubles down on AI as WorkBuddy tops China in AI usage — Download the Zhidx app for an exclusive first look — November 18, 2026, 18:51:18, from Beijing
Zhidx reported on November 18 that Tencent released its financial results for the third quarter (Q3) of 2026 today. As of September 30, 2026, Tencent’s Q3 total revenue was RMB 207.336 billion, up 8% year on year, while net profit was RMB 65.139 billion, up 0.3% year on year.
In terms of spending, Tencent’s Q3 capital expenditure was RMB 81.433 billion, up approximately 527.8% year on year. R&D expenditure reached RMB 32.362 billion, up 41.8% year on year, while cumulative R&D spending in the first three quarters totaled RMB 82.182 billion, with AI-related R&D investment continuing to rise.
▲Tencent’s quarterly revenue and net profit (Graphic by Zhidx)
The financial report showed that Tencent recorded free cash flow of negative RMB 69.1 billion this quarter. Excluding prepayments for computing-power purchases, Tencent’s free cash flow would have been positive at RMB 16.9 billion. This indicates that Tencent made RMB 86 billion in prepayments for computing power.
At the same time, revenue from its computing-power leasing business is also growing. Tencent President Martin Lau disclosed during the earnings call that revenue from Tencent Cloud’s GPU leasing services increased by more than 30% year on year in the third quarter.
Of Tencent’s total revenue, its ToB business—FinTech and Business Services—generated RMB 63.655 billion, up 9% year on year. Growth in Business Services revenue was driven primarily by a 35% year-on-year increase in Tencent Cloud services revenue, although the specific revenue figure for Tencent Cloud was not disclosed.
Sales and marketing expenses totaled RMB 12.144 billion, up 6% year on year, mainly due to increased promotional investment in AI-native products such as WorkBuddy.
▲Revenue share of Tencent’s business segments (Graphic by Zhidx)
Notably, Tencent’s Q3 non-IFRS operating profit declined 6% year on year to RMB 68.1 billion. Excluding its “new AI products”—Hy, Yuanbao, CodeBuddy, WorkBuddy and Xiaowei—non-IFRS operating profit would have been RMB 83.3 billion, up 8% year on year.
This means that the new AI products reduced Tencent’s non-IFRS operating profit by approximately RMB 15.2 billion this quarter, with the drag widening further from Q2.
According to the financial report, as of early November, WorkBuddy had more than 10 million daily active users and averaged over 100 trillion Token calls per day. Measured by users’ total daily consumption of large-language-model Tokens, WorkBuddy has become the most-used AI product in China.
In addition, the daily active user count (DAU) of WeChat AI assistant Xiaowei has reached 30 million. Tencent President Martin Lau said during the earnings call that Xiaowei would eventually become as powerful as Muse and would cover all WeChat users.
“First, this is indeed affecting everyone. The amount of computing power actually available has decreased, affecting both leased and self-built computing capacity.
Second, training will be affected. Tencent is currently highly efficient at training with Hy4 precision from Hunyuan, but the next generation of Hunyuan models may take longer to launch.
At the same time, over the long term, domestic supply chains will step in, and more Chinese companies will rely on domestic supply chains and domestically produced chips.”
I. Cloud services revenue rises 35%, while WorkBuddy revenue quintuples
Regarding foundation models, Hunyuan Hy4 was released on November 4. Compared with the preview version, it expanded the scale of both pre-training and post-training, introduced visual-understanding capabilities for the first time, and supports task execution through both graphical interfaces and command lines. At launch, it ranked first among open-source models worldwide and first among Chinese models on Agent Arena.
Hy4 flash, released on November 16, has 224B total parameters but only 12B active parameters, with overall performance slightly surpassing the larger Hy4 preview. According to Martin Lau, Hy4 flash exceeded 50 trillion Token calls on its first day online and topped OpenRouter’s daily LLM ranking.
Regarding AI applications, Tencent disclosed in its Q3 financial report that WorkBuddy has become the most-used AI product in China, as measured by users’ total daily consumption of large-language-model Tokens. As of early November, WorkBuddy had more than 10 million daily active users and averaged over 100 trillion Token calls per day.
Martin Lau added during the earnings call that WorkBuddy’s annualized revenue over the past seven days exceeded US$500 million, equivalent to approximately RMB 3.334 billion, more than twice the revenue of Tencent Meeting. Its third-quarter revenue was more than five times that of the second quarter.
Alongside this strong growth, Martin Lau also mentioned the sluggish performance of the Yuanbao business, saying that Tencent’s previous investment in Yuanbao had not met expectations and that the company had reduced its spending on the product.
In the personal AI assistant segment, WeChat AI assistant Xiaowei began limited gray-box testing in June this year. By Q3, Tencent had further expanded Xiaowei’s testing coverage and application scenarios. Martin Lau said Xiaowei’s daily active user count (DAU) had reached 30 million. However, Xiaowei is still in the gray-box testing stage.
The financial report showed that Tencent Cloud services revenue increased 35% year on year in the third quarter, with financial returns coming primarily from WorkBuddy, Model as a Service (MaaS) and GPU leasing.
WorkBuddy’s third-quarter revenue was more than five times that of the second quarter, driven mainly by growth in paying users and initial adoption by enterprise customers; MaaS revenue more than doubled quarter on quarter.
The GPU leasing business also maintained its growth. Martin Lau revealed that Tencent’s GPU leasing service revenue increased by more than 30% year on year in the third quarter and also grew quarter on quarter, mainly driven by rising enterprise inference demand and improved computing-power scheduling efficiency.
Tencent’s general and administrative expenses rose 29% year on year in the third quarter, primarily due to increased AI-related R&D investment. Selling and marketing expenses also increased 6% year on year, including promotional campaigns for AI-native products such as WorkBuddy.
Meanwhile, Tencent significantly increased its computing-power procurement. The financial report disclosed that third-quarter operating cash flow included substantial prepayments for computing-power procurement to support upgrades to the Hy series of models, WorkBuddy’s inference demand, and the development of AI capabilities across various products and services, while also meeting external customers’ demand for cloud services.
During the earnings call, Martin Lau said the latest tightening of restrictions had reduced the computing power actually available and would also affect model training.
Tencent Chief Strategy Officer and Group Vice President James added that restrictions on advanced semiconductor imports would not disrupt Tencent Cloud’s revenue growth. The models currently deployed in WorkBuddy and MaaS can all run on unrestricted domestically produced chips.
Furthermore, the fact that Hy4 Flash is 95% cheaper than Hy4 shows that Tencent has also achieved a corresponding increase in the number of tokens that can be generated with a fixed amount of computing power or GPU capacity.
II. WeChat Monthly Active Users Reach 1.449 Billion as Tencent’s Q3 Advertising Revenue Surges 19%
As of September 30, the combined monthly active account count for Weixin and WeChat reached 1.449 billion, up 2% year on year; QQ’s mobile monthly active account count reached 533 million, up 3% year on year. The average daily number of paid value-added service subscriptions was 253 million, down 5% year on year.
In gaming, Tencent’s value-added services revenue totaled RMB 98.326 billion in the third quarter, up 3% year on year. Of this, domestic game revenue was RMB 47.6 billion, up 11% year on year; international game revenue was RMB 18.9 billion, down 9%, or 4% at constant exchange rates; and social networks revenue was RMB 31.8 billion, down 1%.
Growth in domestic games was driven mainly by titles including Delta Force, Valorant: Energy Operation and Honor of Kings. Both Delta Force and Valorant: Energy Operation reached record highs in daily active accounts.
Beyond gaming, advertising was one of Tencent’s standout businesses this quarter.
In the third quarter, Tencent’s marketing services revenue reached RMB 43.049 billion, up 19% year on year. Tencent said clients in major industries increased their marketing spending on its platforms, with particularly notable growth in the consumer goods and local lifestyle services sectors.
On November 11, 2026, at the General Office LinkTime event marking Tencent’s 28th anniversary, Tencent’s management described the AI wave as a matter of corporate “survival” and said: “Every team in every core business must start a new venture for the AI era, shed its burdens, and embrace the new age of AI.”
As demand for model training and inference continues to grow, access to computing-power resources and cost control are becoming increasingly important. Against the backdrop of recently tightening policies related to computing power in particular, companies face greater uncertainty in securing computing resources, and the supply of computing power is no longer simply a matter of how much money is invested.
“Special Notice: The above content, including any videos, images or audio, was uploaded and published by users of Dafeng Hao, an independent media platform operated by Phoenix New Media. This platform merely provides information-storage services.
Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services.”
AI Trends