During its second-quarter earnings call for 2026, Tencent Holdings (00700) executives detailed the company's strategic approach to artificial intelligence, balancing significant investment with a clear path to monetization and a flexible policy on share repurchases.
Chairman and CEO Ma Huateng stated that as the company entered the third quarter, it had achieved major progress across the intelligence, application, and infrastructure layers in building an AI-powered new version of Tencent. He noted that the production version of the Hunyuan 3 model provides users with a powerful and cost-effective large model, laying the foundation for future state-of-the-art capabilities. On the application front, the WorkBuddy AI productivity service and CodeBuddy AI coding tool are experiencing breakthrough user growth and are currently clear leaders in their respective fields in China.
Chief Strategy Officer James Mitchell explained that while the company could generate immediate returns by renting out computing power to third parties—a common practice among cloud providers—Tencent is pursuing a larger strategy. The company is allocating a very significant portion of its new computing capacity to developing its own large models with the goal of reaching an industry-leading level, while simultaneously deploying and popularizing its self-developed AI applications to achieve market leadership in China. The long-term vision is to convert the superior intelligence from these leading models and applications into higher economic returns, for example, by selling tokens through WorkBuddy.
Regarding share repurchases, management stated that capital allocation would be dynamic and reflective of the surrounding environment. If capital expenditure is determined to offer superior returns—such as increasing computing resources to build models and then leasing those resources—the company will direct more cash toward capex and potentially reduce the cash used for buybacks, though this will remain a dynamic situation.
Strategic vision and financial fundamentals
Ma Huateng introduced the company's core strategy by highlighting the "AI-powered new Tencent" and its progress in intelligence, application, and infrastructure. He noted that in the second quarter, total revenue reached RMB 205 billion, an 11% year-on-year increase. Gross profit was RMB 118 billion, up 13%, while non-IFRS operating profit excluding new AI products was RMB 86 billion, up 19%. Key services showed robust performance, with WeChat and WeChat combined monthly active users reaching 1.4 billion. The new game Legend of the Locke Kingdom: World ranked first among all new game releases in China for the year by average daily active users and total revenue. WorkBuddy ranked first in China for AI productivity services based on monthly interactions.
AI strategy and investment justification
President Martin Lau elaborated on the AI strategy, explaining that Tencent's existing businesses, with their inherent moats and AI empowerment, are growing steadily. Network effects, supply chain depth, IP, low churn rates, regulatory requirements, and proprietary data form the foundation of these moats. Beyond this, AI is being deployed to improve returns in areas like gaming and advertising. The new AI initiatives are building on a strong foundation, including the significantly improved Hunyuan 3 base model and the market-leading WorkBuddy and CodeBuddy applications. Lau sees increasing potential for attractive financial returns from these differentiated offerings, noting that the company feels comfortable with large-scale AI investments due to significant upside potential and clear downside protection. In the worst-case scenario, the company could rent out its AI infrastructure via Tencent Cloud at cost-recovery or better prices.
Hunyuan model and application development
Mitchell discussed the success of the Hunyuan 3 production version, which has seen a six-fold increase in total daily token usage across all channels compared to the preview version. The model has consistently ranked among the top three globally on OpenRouter's token usage rankings. The company is accelerating model improvements through a newly built rapid model iteration system and is training a larger parameter model, Hunyuan 4, expected for release later in the year. The strategic rationale for investing in proprietary models is the ability to achieve better unit economics, more innovative features, and greater access to the value of intelligence through co-design across applications, models, and computing infrastructure.
Application layer: WorkBuddy and Xiaowei
WorkBuddy and CodeBuddy are seeing breakthrough success in capabilities and user growth, positioning them as China's leading productivity services based on monthly interactions. The services are achieving high retention rates and user willingness to pay. On the consumer side, Tencent previewed "Xiaowei," an embedded, context-aware agent AI experience within WeChat. Xiaowei leverages WeChat's social graph, knowledge graph, merchant information, and payment capabilities to help users navigate and gain insights from the WeChat ecosystem. It can also help users discover products, make purchase decisions, and place orders, laying the groundwork for agent-to-agent transaction loops. The company is focused on upgrading Xiaowei's capabilities and expanding its service integration in a phased rollout.
Business segment performance
Mitchell reported that total revenue grew 11% in the second quarter, with key segments showing strong performance. Domestic games revenue increased 17%, driven by titles like Delta Force, Valorant, and Legend of the Locke Kingdom: World. Marketing services revenue grew 22% to RMB 44 billion, supported by higher CPMs and impressions. The financial technology and business services segment saw revenue rise 9% to RMB 60 billion, with cloud revenue growth accelerating from mid-teens in the first quarter to low-twenties in the second quarter, driven by AI-related demand, international expansion, and increased usage of general cloud services.
Financial results and capital expenditure
CFO John Lo reported that total revenue for the second quarter of 2026 was RMB 204.8 billion, an 11% year-on-year increase. Gross profit was RMB 118.4 billion, up 13%. Non-IFRS operating profit was RMB 75.6 billion, up 9%, while excluding new AI products, non-IFRS operating profit was RMB 86.1 billion, up 19%. Capital expenditure reached RMB 51.8 billion, a 190% year-on-year increase, as the company accelerated investment in AI infrastructure to support model enhancement, inference demand, and growing external cloud demand. Free cash flow was negative RMB 13.8 billion, reflecting the large-scale AI infrastructure investment and prepayments. The net cash position was RMB 58.2 billion, down from RMB 146.9 billion at the end of the first quarter, primarily due to capital expenditure payments and dividend payments.
Q&A: Capital allocation, ROI, and competitive dynamics
In response to questions about the massive capital expenditure, management explained that the company is investing in a larger strategic vision rather than pursuing short-term rental returns. The investment is divided into two parts: one for the existing, high-quality business, and another for the new AI-native business. The latter is a one-time investment in computing resources to build the AI business, with clear upside potential given the strong performance of the models and new applications. The company noted that for some prepayments made a few months ago, it could sell the computing capacity today at a profit margin of over 30%.
Regarding WorkBuddy's competitive positioning, Martin Lau clarified that it is a new platform designed to solve all productivity needs for office workers. It will host multiple models and skills from various developers, with Hunyuan being one of the primary models but not the only one. The platform's core goal is to solve user problems effectively and efficiently.
On the development of Xiaowei, management dismissed concerns about it cannibalizing existing transaction volumes or advertising inventory. They believe that making the WeChat ecosystem more intelligent will significantly expand its value, similar to how WeChat amplified QQ's value in the mobile era. Xiaowei is designed to enable AI-powered transactions and content discovery, creating a new and enhanced user experience that will unlock substantial value over time.
Addressing questions about the profitability of the AI cloud business, the company noted that domestic token manufacturing costs are extremely low, allowing token businesses to have positive gross margins even at low prices. The gross margin for paid users of WorkBuddy and model-as-a-service is already comparable to the overall Tencent Cloud margin. The company also highlighted that the competitive pricing environment in China's cloud market has improved recently, with Tencent having raised its prices in May.
When asked about the future of the Hunyuan 4 model, management stated that it will be a larger and more capable model than Hunyuan 3, but it will follow the same principles of focusing on real-world usefulness and cost efficiency. The company plans to continue iterating toward state-of-the-art performance with future models like Hunyuan 5, while offering a range of models at different sizes and cost efficiencies.
Regarding the allocation of AI-related spending, Martin Lau confirmed that the investment strategy is flexible and dynamic. The company will maintain prudent spending but will increase investment if it sees clear signs of explosive growth opportunities. The overall spending is controlled within a fixed percentage range of the group's profit, but can be increased if the business can pre-empt huge returns. He emphasized that the company has a clear downside protection strategy: even if it stops developing its own applications, simply leasing out the computing power can generate profits, allowing Tencent to invest for the long term.
James Mitchell added that the company dynamically adjusts the priority of funding allocation within a total budget cap. He noted a significant shift in the allocation of AI spending from the first quarter to the second quarter, with more funds being directed toward the rapidly growing WorkBuddy and away from other products in the AI product line. This dynamic adjustment mechanism will continue to be implemented.
In response to questions about the priority of computing resources for Xiaowei, Martin Lau clarified that the ongoing annual budget for Xiaowei is lower than the normal investment scale for the previous "Yuanbao" product, making the overall cost completely controllable. He expects the business to quickly cover its investment costs as the product experience improves. While the company will not provide precise quantitative targets for profitability, it maintains a disciplined management style with a cap on investment. However, it will increase investment if it believes it can build a large, long-term profitable AI business, always with the safety net of being able to lease out idle computing power for immediate returns.