Tencent Q2: AI Investment Reshapes Income Statement, Capex Surges 176%

Tencent's Q2 2026 earnings show quarterly total revenue of RMB 204.8 billion, up 11% year-over-year; Non-IFRS operating profit was RMB 75.6 billion, up 9%, with operating margin down from 38% to 37% in the same period last year. Growth remains, but the cost of AI investment has become evident.

Capital expenditure was RMB 52.8 billion, up 176% year-over-year, about 2.8 times the same period last year, exceeding market expectations; free cash flow turned negative to -RMB 13.8 billion. Management explained that the negative cash flow was mainly due to large prepayments of approximately RMB 51.4 billion for AI compute procurement, and after stripping these out, free cash flow was RMB 37.6 billion. Quarterly operating cash flow was RMB 52.7 billion, indicating the core business is still generating cash.

The impact of AI investment on profits can be quantified: after stripping out the impact of new AI products such as the Hy series models, Yuanbao, CodeBuddy, WorkBuddy, and Xiaowei, operating profit was RMB 86.1 billion, up 19% year-over-year, with operating margin rising from 39% to 42%. The difference of approximately RMB 10.5 billion between the two sets of data roughly corresponds to the drag on profits from the AI business this quarter.

Pony Ma proposed a three-layer framework in the earnings report: 'Intelligence, Applications, and Infrastructure.' The intelligence layer corresponds to model capabilities; Hy3's official version was released in July and has consistently ranked in the global top three by OpenRouter token consumption since launch. The application layer corresponds to product deployment; WorkBuddy's PC monthly visits reached 20.97 million in June, ranking first in China, surpassing the combined total of ByteDance's Trae and Alibaba's QoderWork.

In compute allocation, Tencent has chosen a different path from some cloud service providers. Chief Strategy Officer James Mitchell stated that current compute demand is strong and rental prices are rising, so it would be possible to rent out compute to third parties to recover depreciation in the short term, but 'Tencent is playing a completely different game.' New compute is prioritized for proprietary model training and proprietary AI application deployment.

Martin Lau added that the priority for compute usage is 'first build proprietary models, then develop applications, and finally rent out externally.' Some compute orders already placed could be sold externally for a profit of more than 30% above procurement cost, but Tencent is forgoing short-term arbitrage to bet on long-term returns.

All three major revenue segments recorded positive growth this quarter. Value-added services revenue was RMB 98.4 billion, up 8% year-over-year, with domestic games revenue of RMB 47.3 billion, up 17%, driven mainly by four titles: Delta Force, VALORANT, VALORANT: Source Action, and洛克王国：世界 (Roco Kingdom: World).

Marketing services revenue was RMB 43.6 billion, up 22% year-over-year, the highest growth among the three segments, driven by AI-optimized ad recommendation models, the upgraded intelligent delivery product matrix Tencent Marketing AIM+, and the integration of closed-loop marketing capabilities within the WeChat ecosystem.

Fintech and business services revenue was RMB 60.3 billion, up 9% year-over-year, with enterprise services growth mainly from cloud services, benefiting from increased AI-related service demand, international expansion, and improved pricing environment. However, cloud revenue growth has not yet matched the pace of capital expenditure growth, and monetization of compute investment is still in early stages.

WorkBuddy is Tencent's most heavily invested product in the AI application layer, with rapid user growth since its launch in March 2026. According to Analysys, its PC monthly visits reached 20.97 million in June, ranking first in China. The earnings report mentioned that WorkBuddy has achieved rapid user growth and healthy user retention, with strong willingness to pay among users.

James Mitchell disclosed that the gross margin of WorkBuddy's paying users and Tencent's model services 'can already rival the overall gross margin of Tencent Cloud.' He also noted that domestic token prices are low, but token production costs 'are also extremely low, far lower than the general perception and estimates,' so the token business can maintain positive gross margins even in a low-price environment.

However, WorkBuddy's overall gross margin is lower than the paying segment because it includes a large number of free users, as Tencent subsidizes market share growth with free quotas. Current profitability signals mainly come from the paying user segment, and overall profitability still requires user scale expansion and improved paid conversion rates.

Tencent's AI investment logic is strategically sound: cash flow from existing businesses supports AI infrastructure construction, and then long-term returns are obtained through leadership in model capabilities and application products. However, there are several risks to track between investment and returns.

First, the intensity and sustainability of capital expenditure. The RMB 52.8 billion far exceeded market expectations, and management hinted that high intensity will continue. Martin Lau admitted that the current AI investment strategy is 'flexible and dynamic,' and they will remain prudent, but 'if we clearly see explosive growth opportunities, we will increase investment,' implying that the peak of capital expenditure may not yet have arrived. Second, the competitive landscape: WorkBuddy is temporarily ahead, but competitors like ByteDance's Trae and Alibaba's QoderWork are iterating rapidly; WeChat's AI assistant 'Xiaowei' is still in small-scale gray testing.

Third, the volatility of free cash flow. Management provided the metric of 'RMB 37.6 billion after stripping prepayments' to alleviate market concerns, but negative free cash flow itself still needs explanation. As of the end of the period, Tencent held total cash of RMB 511.2 billion and the fair value of its listed investment portfolio was RMB 487.2 billion, providing ample financial flexibility. Going forward, we need to track the revenue contribution share of AI products, WorkBuddy's paid conversion rate, marginal changes in capital expenditure, and whether cloud revenue growth can positively match compute investment.