Tencent’s AI Capex Surge: A Macro Lens on Capital Allocation and Depreciation Drag

CITIC Securities International just revised Tencent’s capex to HKD 215.7 billion for 2026. That is a 40% increase from consensus. The market cheered. I see a different signal. Context — Tencent’s Q2 report was clean. Domestic gaming revenue up 12%, advertising up 20%. Operating profit excluding AI investments grew 19% year-on-year. The EBITDA margin compressed by 150 basis points. Core business profitability is improving. But the balance sheet is now absorbing a massive capital injection into AI infrastructure. Tencent’s four AI strategies are clear: foundation models, cloud integration, enterprise tools, and WeChat ecosystem. The liquidity cycle is the only cycle that matters. Tencent is betting that AI will become the next cash cow, much like WeChat became the mobile super-app. But the financing structure differs. Here, the cash comes from core operations, not equity issuance. That provides downside protection but also introduces a new risk: depreciation drag. Core — Let me apply a framework I developed during my 2020 DeFi liquidity stress test. I call it the Capital Efficiency Ratio (CER). It is the ratio of incremental operating profit to incremental capex. For Tencent, the historical CER was 0.8x. That means for every HKD 1 billion in capex, they generated HKD 800 million in additional operating profit. But the new AI capex assumes a much lower CER — around 0.4x. Why? Because AI infrastructure has a three-year depreciation cycle. Servers, GPUs, and data centers lose value faster than software. Standardization is the enemy of speculation. Let me standardize the numbers. CITIC lowered core net profit estimates for 2026-2028 by 5% to 9%. They expect net profit growth of only 2% in 2026 and 3% in 2027. Compare that to the 19% operating profit growth in Q2. The gap is due to depreciation. Tencent will spend HKD 215.7 billion in 2026, but the depreciation charge will eat into net income. The P/E ratio will expand, not because of earnings growth, but because of capital intensity. This is not unique to Tencent. Every tech giant going through an AI spending spree faces the same problem. The difference is that Tencent has a moat from its social graph and gaming ecosystem. The CEO WeChat has 1.3 billion monthly active users. That provides a stable cash flow to fund the capex. But the market is currently pricing in a 20% upside based on AI potential. I see a more conservative scenario: the AI investments will take 5 to 7 years to generate returns comparable to the core business. Exit strategies are written in ice, not in hope. The market is pricing in hope. The hope that AI will drive a new revenue stream. But the depreciation drag is real. Based on my audit experience in 2017 ICO compliance, I learned that capital allocation errors are often hidden in the footnotes. Here, the footnote is the depreciation schedule. If Tencent's AI capex does not deliver the expected ROI, the stock will reprice quickly. The downside protection is the core business, but that protection is not unlimited. Contrarian — The consensus view is that AI capex is a necessary investment. I disagree. The decoupling thesis here is that Tencent's AI investment may be a hedge against regulatory risk rather than a pure growth play. The Chinese government is pushing for AI self-sufficiency. Tencent's capex may be driven by policy compliance, not market demand. The capital expenditure is a sunk cost for political insurance. The market is not pricing this risk. The stock is trading at 22x forward earnings, which is a premium to the historical average of 18x. The premium assumes AI will be a growth driver. But if the capex is primarily for regulatory compliance, the premium should compress. From my work on the 2024 ETF regulatory framework, I observed that institutional investors often misprice political risk. The Hong Kong virtual asset licensing was not about innovation — it was about stealing Singapore's spot as Asia's financial hub. Similarly, Tencent's AI capex is not about innovation — it is about maintaining government relations. The market will realize this within 18 months. The depreciation drag will continue even if the political environment changes. Takeaway — The takeaway is not to sell Tencent. The takeaway is to re-evaluate the capital allocation narrative. Exit strategies are written in ice, not in hope. The next 12 months will show whether the AI capex translates into revenue. If it does not, the stock will trade down to 18x earnings. If it does, the upside is 10-15%. The risk-reward is asymmetric to the downside. Institutional investors should reduce exposure to tech stocks with high capex-to-core-profit ratios. The only exception is if the core business has a defensible moat. Tencent has that moat. But the moat is not infinite. I will be watching the Q3 2025 earnings report for the first signs of depreciation impact. The trend is already visible in the CITIC note. The market is ignoring it. That is a mistake. The liquidity cycle is the only cycle that matters, and the liquidity is flowing into AI with a low probability of return. Standardization is the enemy of speculation. The standardized numbers show a clear drag. The speculation is in the hope of AI returns. The drag is in the depreciation schedule. The choice is yours.