Tencent nearly tripled its capital spending in the second quarter of 2026 to accelerate its AI buildout, reporting revenue of 204.8 billion yuan (US$30.4 billion) for the quarter ending in June — up 11% year on year from 184.5 billion yuan and ahead of the average analyst estimate of 202.8 billion yuan compiled by Bloomberg1. The company reported unaudited results on August 122.
Capex surged 176%, driven by accelerating investments in computing power and models that management said helped drive the revenue beat. Fintech and enterprise services revenue reached 60.3 billion yuan, up 9% year on year.
On the earnings call, Chairman and CEO Pony Ma outlined a three-layer AI strategy spanning intelligence, applications, and infrastructure.
The intelligence layer is anchored by the Hunyuan Hy3 model, which was released recently and rapidly integrated into products including Yuanbao. Hy3 is now available globally through an API and through Tencent's WorkBuddy international version. Hy4 is planned for near-term release.
The application layer showed the strongest traction, according to the company. WorkBuddy, which Tencent calls China's leading AI-native office agent, posted rapid user growth with healthy retention across iOS, Windows, Android, and HarmonyOS. Tencent cited engineering capability, model selection, and a skills library as WorkBuddy's strengths. Martin Lau described WorkBuddy as a flexible workspace for agentic AI aimed at office productivity needs and one-person companies, calling it a new platform rather than enterprise software. WorkBuddy uses a harness layer that leverages different models and a growing library of third-party skills; Hy will be one of the models offered and will serve as the primary model if it solves problems cost-effectively, but not the only one. The broader application portfolio has expanded to include Marvis and QClaw, and CodeBuddy achieved what the company described as breakthrough user growth.
On the infrastructure layer, Tencent significantly increased compute procurement to convert model and application usage into revenue. Chief Strategy Officer James Mitchell said that given surging demand and rising lease prices, Tencent could almost immediately recover depreciation costs by renting compute to third parties, but chose instead to direct most new compute toward training its own models and deploying its own AI applications. Martin Lau added that compute ordered or prepaid months ago could now be resold at a margin above 30% over the original purchase price. Management framed the AI-native capex as an upfront investment for model training, future inference demand, and Tencent Cloud's AI capacity.
Management described the company's business as comprising two parts: the existing business and a new AI-native business built on self-developed models, new applications, and associated compute. Tencent said it believes superior intelligence converts to superior economics over the long run, including through token sales via WorkBuddy. The quarter was framed by management as the beginning of an AI-empowered Tencent.
Beyond AI, Tencent said marketing services grew, several new games performed well, and WeChat Channels video views rose rapidly.
ANALYSIS The decision to absorb compute internally rather than resell at a 30%-plus margin signals that Tencent is prioritizing model and application competitiveness over near-term infrastructure revenue. The multi-model harness architecture of WorkBuddy — where Hunyuan is primary but not exclusive — positions the product as a platform play distinct from single-model assistants, tying application stickiness to the skills ecosystem rather than to any one foundation model.