Tencent's aggressive positioning in AI infrastructure has come into sharper focus following its latest investor communications.
According to a recent research note from HSBC, Tencent's Chief Strategy Officer James Mitchell disclosed during an investor roadshow call that the company prepaid over RMB 50 billion in the second quarter to secure supply of current and next-generation memory chips. Mitchell also highlighted the growing monetization potential of the Harness product suite, noting that gross margins for paying users' inference workloads have now matched the company's MaaS business, while paying users consume tokens at more than double the rate of free users.
The comments were made during HSBC's non-deal roadshow (NDR) conference call held on September 3, 2026. HSBC subsequently maintained its Buy rating on Tencent with a target price of HK$655, implying an approximately 47.9% upside from the current share price of HK$442.80.
Where the capital intensity stands now
HSBC noted that second-quarter capital expenditure may represent the new steady-state operating level, though management acknowledged room for fluctuation around this baseline. Management indicated that the major leap in capital intensity was completed in the second quarter, with prepayments potentially extending into the third quarter before normalizing from the fourth quarter onward.
A prepayment scale rarely seen, described as once-in-five-years
Per the HSBC research report, Tencent's prepayment of over RMB 50 billion in Q2 was primarily directed at locking in current and next-generation memory chips at attractive prices. Management characterized the procurement as a partially one-off expense that may occur roughly once every five years, aimed at addressing memory chip supply bottlenecks.
On the cadence of capital expenditure, management stated that Q2 spending levels can be viewed as the new normal baseline, while stressing that there remains room for upward or downward adjustments. The company will continue chip purchases through the remainder of this year and into 2027, but the primary capital intensity surge was completed in Q2. HSBC projects Tencent's full-year 2026 capital expenditure at approximately RMB 212.4 billion, a sharp increase from RMB 112.7 billion in 2025.
Regarding AI product expenses, management noted that the main cost drivers in Q2 shifted from marketing spend on Yuanbao in Q1 to operating expenses for the Hunyuan large model (HY) and Harness products such as WorkBuddy and CodeBuddy. In contrast, Xiaowei's operational costs are expected to be significantly lower than HY or WorkBuddy, attributed to its use of the lightweight WeLM model, which demands lower computational resources.
Harness monetization gains traction as paying user margins match MaaS
On the priority ranking of AI monetization pathways, management acknowledged that MaaS currently offers the highest immediate returns, with gross margins around 40%, driven by GPU shortages and large-scale training demand. However, Tencent has chosen to prioritize long-term resource allocation toward Harness products and Hunyuan model training over the more near-term lucrative MaaS segment.
Management's rationale is that MaaS gross margins will face pressure in the future as market demand shifts from training to inference, and as model vendors increasingly build their own computing infrastructure, reducing reliance on cloud providers as distribution channels. In contrast, Harness revenue growth and gross margins present upside potential, supported by rising inference demand, stronger user stickiness through improved task history memory features, and continued conversion of free users to paying subscribers.
The HSBC report cited management saying that Harness paying users' inference gross margins have already reached parity with MaaS, and that paying users consume tokens at two times or more the rate of free users, demonstrating robust user-tier monetization capabilities.
The pursuit of SOTA models and the CodeBuddy gap driving faster self-developed model iteration
On AI strategy, management explained the rationale behind Tencent's current pursuit of state-of-the-art (SOTA) models. Management conceded that team restructuring had delayed model R&D progress by six to nine months, but a new unified reporting structure has accelerated the model release cadence to once every two months.
At the product level, WorkBuddy can mitigate the risk of third-party model unavailability by guiding users to Hunyuan 3 (HY3), but CodeBuddy previously lacked robust support from a strong self-developed code model. The Hunyuan 4 preview (HY4 preview) delivers substantial improvements in code capabilities, enabling it to handle a portion of CodeBuddy's code-related requests and bridge this gap. Management also noted that owning proprietary SOTA models will help improve the long-term margin structure.
Comments