Tencent CFO: RMB 50 billion prepayment secured memory chips in Q2; Harness's gross margin for paid users is now comparable to MaaS and its consumption rate is more than twice that of free users.

Tencent's massive bet on AI infrastructure is coming to light.

According to a recent research report by HSBC, Tencent's Chief Strategy Officer, James Mitchell, disclosed in an investor roadshow call that the company prepaid more than RMB 50 billion in the second quarter to lock in existing and next-generation memory chips. He also revealed that Harness 's monetization potential is becoming increasingly apparent, with the gross margin of inference for paid users now on par with the MaaS business, and paid users' token consumption rate being more than twice that of free users.

The above statement comes from HSBC's NDR roadshow conference call held on September 3, 2026. In its subsequent research report, HSBC maintained its buy rating on Tencent, with an unchanged target price of HK$655, implying an upside of approximately 47.9% from the current share price of HK$442.80.

HSBC noted that capital expenditure in the second quarter may have become the new normal, but there is still room for fluctuation. Management stated that the surge in capital intensity was completed in the second quarter, and advance payments may continue into the third quarter, before normalizing from the fourth quarter onwards.

The scale of chip prepayments is unprecedented, described by management as a "once-in-five-years" event.

According to a HSBC research report, Tencent's prepayment of over RMB 50 billion in the second quarter was primarily used to secure current and next-generation memory chips at attractive prices. Management characterized this procurement as a partial one-off expenditure, or an event that occurs every five years, aimed at addressing memory chip supply bottlenecks.

Regarding the pace of capital expenditure, management stated that the spending level in the second quarter can be considered a new benchmark, but emphasized that there is still room for fluctuation. The company will continue to purchase chips for the remainder of this year and into 2027, but the major leap in capital intensity was completed in the second quarter. HSBC expects Tencent's total capital expenditure for 2026 to be approximately RMB 212.4 billion, a significant increase from RMB 112.7 billion in 2025.

Regarding the costs of new AI products, management pointed out that the main driver in the second quarter shifted from the marketing expenses of Yuanbao in the first quarter to the operating expenses of Harness products such as Hunyuan Large Model (HY) and WorkBuddy and CodeBuddy. In contrast, Xiaowei's operating costs will be significantly lower than HY or WorkBuddy because it uses the lightweight model WeLM, which has lower computational cost requirements.

Harness's monetization potential is becoming increasingly apparent, with gross margins for paid users catching up with MaaS.

Regarding the priority of AI monetization paths, management admitted that MaaS currently offers the highest return on investment—driven by GPU shortages and the demand for large-scale training, MaaS's gross margin is currently around 40%. However, Tencent has chosen to prioritize long-term resources for Harness products and mixed-model training, rather than MaaS, which offers more certain short-term returns.

Management's rationale is that as market demand shifts from training to inference, and model vendors gradually build their own computing power infrastructure and reduce their reliance on cloud service providers' distribution channels, MaaS gross margins will face pressure in the future. In contrast, Harness's revenue growth and gross margin have room for improvement, driven by factors including increased inference demand, improved user stickiness (due to improved task history memory functionality), and the continued conversion of free users to paying users.

A HSBC research report, citing management, stated that Harness's gross margin for paid users' inference is comparable to that of MaaS, and that paid users' token consumption rate is twice or more than that of free users, demonstrating a strong user segmentation and monetization capability.

Behind the pursuit of state-of-the-art (SOTA): CodeBuddy's shortcomings force the acceleration of self-developed models

At the AI strategy level, management explained the strategic motivation behind Tencent's current pursuit of state-of-the-art (SOTA) models. Management acknowledged that team restructuring had delayed model development by 6 to 9 months, but the new unified reporting structure has accelerated the model release cycle to once every two months.

At the product level, WorkBuddy can mitigate the risk of third-party model unavailability by directing users to Hybrid 3 (HY3), but CodeBuddy previously lacked robust self-developed code model support. The Hybrid 4 preview version (HY4 preview) significantly improves code capabilities, enabling it to handle some code requests from CodeBuddy, thus filling this gap. Management also pointed out that having a self-developed state-of-the-art (SOTA) model will help improve the profit margin structure in the long term.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.