Aggressive spending on chips and models weighs on earnings even as core businesses remain robust
By Lian Ran
Tencent’s (0700.HK) AI story is entering a more expensive phase.
Revenue rose 11% year on year to 204.8 billion yuan ($30.4 billion) in the second quarter, while non-IFRS operating profit increased 9% to 75.6 billion yuan and non-IFRS net profit attributable to shareholders grew 9% to 68.4 billion yuan.
But strip out the costs and revenue of new AI products including Hunyuan, Yuanbao, CodeBuddy, WorkBuddy and Xiaowei, and non-IFRS operating profit would have risen 19% to 86.1 billion yuan. That implies the new AI businesses had a net negative impact of approximately 10.5 billion yuan on operating profit in the quarter, up from 8.8 billion yuan in the first quarter, increasing the drag on overall profit growth.
More striking was the change in cash flow. Capital expenditure jumped 176% year on year to 52.8 billion yuan, while research and development spending rose 35% to 27.3 billion yuan. With large prepayments for computing capacity, free cash flow turned negative at 13.8 billion yuan. Excluding those prepayments, it remained positive at 37.6 billion yuan.
Why Tencent is spending big
Tencent’s AI strategy has shifted from rebuilding its research organization and integrating AI into products to materially reshaping its financial statements.
Chief Strategy Officer James Mitchell said Tencent could rent out computing capacity and recover depreciation costs quickly, given strong demand. But the company has chosen a different path: directing most new capacity into proprietary LLMs and applications, aiming to convert that intelligence into longer-term, higher-margin returns through channels such as WorkBuddy token sales.
President Martin Lau said Tencent’s core businesses — games, advertising and fintech — remain strong cash generators, with capital expenditure covered by operating cash flow. AI-native businesses, by contrast, require concentrated upfront spending on model training, inference capacity and cloud infrastructure.
Some computing capacity prepaid several months ago could now be resold at profits of more than 30%, Lau said. Capacity is prioritized for Tencent’s own models and products such as WorkBuddy, with excess rented out through Tencent Cloud.
That explains why Tencent is willing to spend more. WorkBuddy’s paid users and its model-as-a-service business have gross margins comparable with Tencent Cloud overall. Cloud revenue growth accelerated from about 18% in the first quarter to roughly 21% in the second, driven by GPU leasing, model services, and token revenue.
Xiaowei: WeChat’s AI experiment
While WorkBuddy is Tencent’s attempt to commercialize AI in productivity, Xiaowei — WeChat’s AI assistant still in beta — represents a larger ambition: turning WeChat from a super-app where users search, click and navigate services themselves into an intelligent agent ecosystem that can understand instructions, coordinate services and complete transactions.
Lau rejected the view that AI agents might merely migrate existing transactions to a costlier channel. He framed it as a re-magnification of WeChat’s ecosystem value.
Tencent’s longer-term vision is an “agent-to-agent transaction loop.” Users, merchants and mini-program providers will all have agents. After a user gives an instruction to Xiaowei, the user-side agent can collaborate directly with merchant-side agents to handle inquiries, filtering, matching, ordering and payment.
This remains long-term. Beyond model capabilities, there are complex issues around permissions, identity, payment security, privacy, service standards and merchant onboarding. Tencent’s approach is phased.
Xiaowei leverages WeLM, a proprietary WeChat model focused on privacy, scenario adaptation and inference efficiency — distinct from Hunyuan, Tencent’s proprietary, in-house large language model family. The latter pursues general capability frontiers; Xiaowei must balance experience with low cost across hundreds of millions of users.
Gaming provides the cash-flow cushion
Tencent’s AI spending is being supported by an acceleration in its most important cash-generating business. Gaming revenue rose 11% year on year to 65.9 billion yuan in the second quarter. Domestic gaming revenue increased 17% to 47.3 billion yuan, accelerating from 6% in the first quarter.
New titles including Delta Force, Valorant and Rock Kingdom: World are adding growth alongside established franchises.
AI is also entering Tencent’s game-development pipeline. Delta Force uses AI agents for performance analysis and Tencent’s Hunyuan 3D model to generate some assets. Peacekeeper Elite has introduced AI NPCs, while Rock Kingdom: World uses a Coach Agent called Scarlett.
The first phase of AI helped produce content faster. The next phase will test whether it can create dynamic content – teammates adjusting to player ability, persistent NPCs, user-generated maps – potentially extending evergreen games’ lifecycles.
AI now has to prove its returns
Tencent’s Hong Kong-listed shares have fallen 26% year-to-date and are down almost 30% from their 52-week high. The market’s concern is is how much Tencent will spend, for how long, and who will ultimately pay for it.
Management has stressed that this is not unlimited spending. AI-related capital expenditure is expected to be concentrated over the next two years, resources will shift toward products showing clear growth, and computing capacity can be rented out if necessary through Tencent Cloud.
Tencent differs from AI companies betting on a single breakout. It uses gaming and advertising cash flow as backstop, cloud for spillover, WorkBuddy for B2B productivity, and Xiaowei for WeChat’s evolution. Whichever track first forms a positive loop of user growth, revenue and cost will draw more resources.
With over 50 billion yuan in capital spending, the shift is clear: from “light R&D, fast iteration” to “heavy computing, long-cycle, ecosystem.”
Profit growth may slow further. Over the next year or two, the market will need to get used to a Tencent that converts more current profits into GPUs, models and next-generation infrastructure.
Whether that trade-off pays off will depend on whether WorkBuddy becomes a genuine productivity platform, Xiaowei can reshape the WeChat ecosystem, and Hunyuan remains competitive as AI technology evolves. Those answers are still unfolding.
Source:
geekparkGO
