Tencent's AI Approach and CapEx Stress Test

Cross-validated by this quarter’s earnings call, Tencent is gradually finding its own path in AI: rather than engaging in direct competition with peers on model parameters, it is leveraging its core strengths—scenario density, ecosystem depth, and data assets—to establish a differentiated competitive position. This approach is internally coherent but comes with clear boundaries.

Currently, two core questions frame the analysis of Tencent: first, what share of the AI-driven incremental opportunity can Tencent realistically capture; and second, to what extent will rising capital expenditures subsequently weigh on reported profits, and how exactly will this pressure manifest?

1. Found its own rhythm, but still not a frontrunner

The pace of catching up at the model layer has been relatively fast. Tencent overhauled its training pipeline in February, and by April had already released its next-generation HunYuan model, achieving an iteration cycle of roughly three months. This tempo further underscores that large models do not face insurmountable technical barriers; whether the gap can narrow depends more on strategic resolve, talent density, and organizational coordination. HunYuan currently holds a stable position in the second tier, with room to move further upward.

More noteworthy is Tencent’s clear strategic prioritization at the model layer. Agents—the connective layer between model capabilities and specific use cases—will be the key focus going forward.

According to OpenRouter data, HunYuan Hy3 preview remains the top daily token consumer even after ending its free trial period, ranking first in tool invocation and second in code generation among细分 capabilities. It should be noted, however, that this data suffers from sample bias—OpenRouter’s user base is predominantly developers, and Tencent itself possesses strong promotional capabilities—so this does not directly equate to market-wide leadership.

What this section truly aims to highlight is the shift in competitive dimensions. Over the past year, leading models have taken turns at the forefront, with no single player able to establish a sustained advantage. This implies that models themselves are transitioning from a moat-like attribute to a baseline entry requirement—the performance gap is increasingly temporary rather than permanent. In the next phase of AI competition, success will hinge on a different set of factors: depth of scenario accumulation, completeness of ecosystem architecture, and usability of data assets.

Under this new framework, Tencent’s inherent strengths stand out. This quarter, WeChat search query volume grew 25% year-over-year, compared with Google’s traditional search query growth of approximately 5%. Mini Programs are being systematically re-engineered into a Skills layer directly callable by Agents, unlocking the potential for systematic reuse of ecosystem assets accumulated over the past decade. WorkBuddy also became China’s top AI-powered workplace Agent by daily active users this quarter. Viewed through this lens, the industry-wide shift from 'competing on models' to 'competing on ecosystems' presents Tencent with a relatively favorable window—assets competitors invested in early are now depreciating, while Tencent’s long-accumulated assets are appreciating in value.

However, execution speed remains a relative weakness for Tencent. Organizational efficiency improved noticeably this quarter—completing model iterations within three months, rapidly embedding Yuanbao into WeChat/QQ/QQ Music, and finalizing the B2B WorkBuddy product suite—but compared with industry leaders, Tencent’s overall pace still lags. This round of adjustments has moved Tencent from 'clearly behind' to 'barely keeping pace,' yet it remains significantly distant from the true frontrunners. Should the industry undergo another leapfrog advancement, market skepticism toward Tencent would likely resurface. A telling detail worth noting is that Pony Ma recently expressed a rare sense of urgency at the latest shareholder meeting, stating verbatim: 'We’ve boarded the ship, stood up, but haven’t sat down yet.' On the flip side, Tencent’s slower pace isn’t entirely negative—it has created a window for internal reorganization and strategic reassessment.

2. Two sides of the same coin: Deep moats, but questionable offensive capability

When examining Tencent’s AI strategy in detail, a common pattern emerges: nearly every business line ultimately seeks its anchor point within the WeChat ecosystem. Yuanbao’s breakthrough relies on WeChat as an entry point; Agent deployment depends on Mini Program service offerings; and most new monetization opportunities must be reallocated within the WeChat domain.

In the short term, this dependency is an advantage. The existing moat is deep enough that new capabilities don’t need to build scenarios from scratch—they can be directly grafted onto established high-frequency touchpoints for monetization. Whether on the consumer side (WeChat, Yuanbao, QQ Music) or the enterprise side (WorkBuddy, CodeBuddy, Tencent Meeting, Tencent Docs), AI integration focuses on 'enhancing capabilities within existing scenarios'—keeping implementation costs manageable, ROI pathways clear, and failure risks limited.

However, over a longer time horizon, this also implies that Tencent’s growth ceiling could be capped by the inherent limitations of the WeChat ecosystem itself. The question—'Will AI disrupt Tencent’s moat?'—is relatively straightforward to answer: most likely not; in fact, it may even be reinforced. However, there is currently no clear answer to how much of the new incremental market created by AI Tencent will be able to capture. This remains a core issue requiring ongoing monitoring.

The current relative weakness of the Yuanbao app serves as a cautionary footnote. In the race to develop standalone consumer-facing AI products, Yuanbao has not outperformed competitors such as Doubao and DeepSeek. This reflects Tencent’s limited organizational capability in 'building new growth from scratch'; its offensive strength remains heavily reliant on its existing ecosystem. Should AI ultimately evolve a completely new 'super-agent entry point' that reshapes the entire mobile internet traffic allocation landscape, it remains uncertain whether Tencent can secure a share commensurate with the scale of its ecosystem.

III. CapEx Stress Test: Increased Investment with Relatively Contained Profit Impact

Tencent’s AI-related spending has been relatively restrained compared to peers, with no aggressive pursuit of large-scale capital expenditure expansion. Its 2025 CapEx stands at RMB 79.2 billion, representing approximately 10.5% of revenue. During this quarter’s earnings call, management guided that investment in HunYuan, Yuanbao, and other new AI products will double in 2026 relative to 2025 levels, funded by profit growth from core businesses.

One strategic choice deserves special mention: for some time now, Tencent has prioritized allocating GPU resources to internal AI initiatives—including HunYuan, WeChat Agent, Yuanbao, ad recommendation systems, and WorkBuddy—rather than rushing to lease significant computing capacity externally via Tencent Cloud. This decision reflects the belief that the ROI from internal use cases exceeds that from selling computing power externally. While this is a positive signal from a long-term value perspective, it delays the near-term revenue realization timeline for its cloud business.

To clearly establish the lower bound of potential profit impact, we can construct a pessimistic stress test scenario. Assuming Tencent maintains an average annual CapEx of RMB 150 billion over the next six years, the steady-state annual financial impact would be roughly as follows.

On the negative side, the primary impact comes from depreciation and amortization (D&A): RMB 150 billion in annual CapEx translates into approximately RMB 150 billion in steady-state D&A. Net of the RMB 46 billion already embedded in the base case, this implies an incremental D&A pressure of roughly RMB 100 billion per year, or about 8–9% of revenue.

On the positive side, the total benefit ranges between RMB 95–140 billion annually. This includes approximately RMB 40–50 billion in net profit contribution from cloud and AI cloud services and high-value subscription revenue expansion (net of associated depreciation); RMB 10–20 billion from AI-driven advertising efficiency gains (higher eCPM plus new ad inventory); RMB 25–35 billion from direct cost savings (25% workforce substitution plus AI-assisted content production); and RMB 20–30 billion from indirect productivity improvements (greater output per employee).

After offsetting both sides, the net steady-state impact falls within a range of -RMB 5 billion to +RMB 40 billion per year—essentially neutral to slightly positive.

However, this conclusion rests on three key assumptions: first, that RMB 150 billion in annual CapEx can effectively translate into corresponding business capabilities (in cloud and subscriptions); second, that cost reductions and efficiency gains can be meaningfully realized; and most critically, third—that the pace of AI monetization aligns with the pace of investment. The last point represents the core risk: if investment outpaces returns, the net impact over the first three to four years will be significantly negative. In other words, the return curve on AI investment is likely to follow a J-shape—initially pressuring profits, before turning mildly positive once steady state is reached.

Another implication of Tencent’s restrained AI investment is that, compared to peers significantly expanding their capital expenditures, it faces substantially less earnings pressure. Even if capex were to rise to RMB 150 billion in 2026, this would still represent a manageable impact relative to its approximately RMB 270 billion non-IFRS net profit base. A forward-looking indicator worth tracking over the long term is the gross margin of its advertising business—down 50 basis points year-over-year this quarter, marking the first visible erosion from AI-related depreciation at the gross margin level. The key observation going forward will be whether the pace of AI monetization can outstrip the pace of depreciation expansion. This signal is critical for identifying the inflection point on the J-curve.

IV. Prudence Is a Double-Edged Sword

The market’s relatively low valuation of Tencent does not stem from operational missteps but rather reflects the current market preference for 'aggressive narratives.' In this environment, investors are more willing to price in imagination, whereas Tencent emphasizes certainty and tends to adopt a follower strategy—traits that inherently struggle to command equivalent valuation premiums. This duality defines Tencent: prudence helps it avoid severe setbacks from a single mismatch during the AI cycle, yet that same prudence becomes a valuation drag under today’s market sentiment.

Perhaps a more suitable valuation framework for Tencent isn’t betting on the elasticity of AI-driven upside, but first clearly quantifying its margin of safety. Even under the assumption that Tencent fails to capture significant incremental market share in this AI wave and merely uses AI to optimize existing operations—effectively stripping away its offensive capabilities and retaining only defensive attributes—its core businesses could still support a relatively solid profit base.

Therefore, three key dimensions truly merit inclusion in the investment thesis. First, AI will not erode but rather reinforce the moats around Tencent’s core businesses—a point largely validated by this quarter’s results (advertising +20%, WeChat search +25%, and video channel watch time +30%). Second, Tencent has a sufficiently robust profit base to sustain its AI investments without falling into financial distress from capex—this quarter’s non-IFRS net profit stood at RMB 67.9 billion (+11%), free cash flow at RMB 56.7 billion, and ending net cash at RMB 146.8 billion, providing ample financial cushion. Third, there’s a still-unconfirmed but attractive asymmetric opportunity: as competition shifts from 'models' to 'ecosystems,' Tencent’s existing assets may undergo revaluation by the market.

High certainty, limited imagination. In this highly uncertain AI race, Tencent may not be the top-performing stock, but it is likely one of the most compelling options in terms of certainty-to-valuation ratio. It may not become the protagonist of this narrative, but in hindsight, it will probably be among the few capable of navigating the entire cycle intact.

Risk disclaimer: The above content represents personal views and theoretical analysis only and does not constitute any investment advice. Please assume all risks yourself.

137K Views