Opportunities and Concerns of Tencent's (00700) 84.7 Billion Capital Expenditure

Is Tencent betting on tomorrow amid the AI wave with todays profits? The latest mid-term performance report of TENCENT for 2026 revealed a capital expenditure figure of 84.72 billion yuan for the first half of the year, quickly drawing strong attention from the market. Notably, the capital expenditure for the second quarter alone reached 52.784 billion yuan, surging 176% year-on-year, exceeding the spending levels for any full year between 2019 and 2023. On the first trading day following the earnings report, Tencents American depositary receipts (ADR) fell by 5.44%; on August 13, its Hong Kong shares opened more than 3% lower and dipped over 4% during the session, with total market capitalization briefly falling below 4 trillion HKD. As of the close on August 14, Tencents shares were priced at 440 HKD each, indicating that its stock price remains in a period of market sentiment adjustment following the semi-annual report data. It is worth mentioning that Tencent, which has consistently held the title of the number one stock in Hong Kong, now has a market capitalization of 400.54 billion HKD (approximately 3.441 trillion RMB), having been surpassed in market value for two consecutive days by the newly listed CXMT Corporation, which has a market capitalization of 3.69 trillion RMB. From maintaining a light-asset operation in consumer internet with NETDRAGON to spending nearly 85 billion yuan on AI infrastructure in just half a year, Tencent finds itself at a critical juncture in its business model transition. The flow of this enormous capital, the clarity of the return path, and the limitations of risks have become focal concerns for the technology industry and capital markets alike. Single-quarter capital expenditure of 52.8 billion: Prioritization of Tencent's investment in computing power Breaking down the capital expenditure of 84.72 billion yuan over six months, purchasing computing power infrastructure stands out as the absolute core destination. According to Tencent's financial report, approximately 70% to 80% of the 52.784 billion yuan capital expenditure in the second quarter was directed toward hardware procurement such as high-performance GPU servers, AI acceleration chips, and networking equipment, used to support the training of large models and AI application inference needs; the remainder covers civil construction for data centers, supporting electromechanical construction, and general cloud infrastructure expansion. Notably, in addition to the on-paper capital expenditure, Tencent also made a prepayment of 51.4 billion yuan for computing power procurement in the second quarter, which was an advance payment made to chip and server manufacturers for capacity reservation. When accounting for this prepaid amount, Tencents actual investment in AI infrastructure for the second quarter exceeded 100 billion yuan. Reporters from the Science and Technology Innovation Board Daily noted that the large investment has directly impacted the cash flow statement. Tencent recorded a free cash flow of -13.8 billion yuan in the second quarter, marking its first negative figure since 2005. Tencents CFO Luo Shuohan explained on the earnings call that the net cash inflow from operating activities was 52.7 billion yuan, which was more than offset by capital expenditure payments of 59.3 billion yuan, media content payments of 5 billion yuan, and lease liability payments of 2.2 billion yuan. Excluding the 51.4 billion yuan prepayment for computing power procurement, the actual free cash flow stood at 37.6 billion yuan, indicating that the cash flow generation capability of its traditional core businesses remains healthy. Regarding this record-setting investment, Tencent's President Liu Chiping explicitly delineated it into two categories during the earnings call: First, expenditure on existing businesses, which still possesses a high cash flow generation capability; Second, capital expenditure for AI-native businesses, aimed at model training, inference needs, and the construction of AI computing and cloud services. He repeatedly emphasized to the market that AI-related capital expenditures are concentrated in the initiation phase of the current and next year, and should not be assumed to maintain the same scale every yearThe model-building part is more inclined towards fixed costs. While it is indeed necessary to secure sufficient computing power, this does not imply that more investments will be required each year. In terms of internal resource allocation, Tencent has a clear prioritization order. Tencent's Chief Strategy Officer James Mitchell revealed that the urgent priority for capital expenditure is to train larger and more powerful large models in the coming months; secondly, to provide inference support for models running behind WorkBuddy, among others; external rental of computing power is ranked last. Internally, Tencent Cloud believes that the Token production of WorkBuddy holds the most enduring economic value, which is why it is currently prioritized for development. Monetization roadmap: Three-tiered approach from models to applications The core market concern always points to one question: Can the hundreds of billions in computing power investment really commensurate commercial returns? On the earnings call, Liu Chiping provided a clear three-tiered advancement logic: first, leverage computing resources to build in-house models, then develop proprietary AI applications, and finally rent out the remaining computing power externally, thereby creating a large-scale, highly profitable AI-native business that generates cash flow. The foundational model is the starting point of the entire path. In July of this year, Tencent officially released the Hy3 version of the large model, with tremendous performance improvements over the preview version, reduced hallucination rates, enhanced long context and agent capabilities, and a daily average Token usage across all channels that grew approximately sixfold compared to the preview version, consistently ranking among the top three in OpenRouters global Token usage chart. Liu Chiping revealed that a larger parameter-scaled Hy4 is set to be released in late 2026, continuing to align with the industrys top levels. Currently, all major companies are engaged in foundational model development, seizing the window to catch up with the first tier, which is central to Tencents capital expenditure usage. As for whether it will yield returns, I believe we need to take a long-term perspective, said Wang Xiangjie, Chief Analyst at Southwest Computers and Overseas Industries, in an interview with reporters from the Science and Technology Innovation Board Daily. The commercialization progress at the application level is currently the most tangible signal of returns. Launched in February of this year, the office AI agent WorkBuddy is the Tencent AI product with the clearest monetization path. Starting July 1, its subscription system has expanded from free + professional version to four tiers, with standard at 99 yuan/month, premium at 199 yuan/month, and flagship at 999 yuan/month. It is now available in over 50 industries, covering areas such as government affairs, production, and life services. James Mitchell disclosed that the gross margin of WorkBuddys paying users is already comparable to Tencent Clouds overall gross margin, while the current overall gross margin is relatively low, mainly due to subsidies for free users lowering the overall level. In Wang Xiangjies view, the value of WorkBuddy extends beyond being a single product. He analyzed for the Science and Technology Innovation Board Daily: For Tencent, WorkBuddy represents a crucial exhibition of its product-level capabilities. Tencent must demonstrate that it retains significant competitiveness in the AI era through the product integration capabilities brought by foundational models. The WeChat ecosystem, with 1.4 billion monthly active users, remains Tencent's most unique long-term advantage. The financial report indicated that by the end of the second quarter, the number of WeChat and WeChat monthly active accounts reached 1.439 billion. Currently, WeChats dedicated AI assistant Xiao Wei has begun greyscale testing and can tap into WeChat's social, knowledge, merchant, and payment ecosystem capabilities. Liu Chiping stated that the investment scale for Xiao Wei is smaller than for the earlier Yuanbao initiative, and the costs are fully controllable, with returns expected to manifest quickly as user experience is enhanced. He likened it to the leap of the mobile internet era: QQ was a communication tool in the PC era, but in the mobile era, WeChat multiplied its value tenfold; in the AI era, WeChat will evolve from being empowered by AI to an AI-first ecosystem, creating incremental value instead of replacing existing business models. Moreover, Tencent has prepared downside protection for its hefty investment. Liu Chiping revealed on the call that the prepaid computing power orders from months ago could currently be turned around for more than a 30% profit. If we were to change our business model today to simply renting out computing power, not only would we not incur losses, but we would actually achieve profits. Therefore, we have always maintained this alternative option, which is why we feel secure. Zhang Xiaolu, Deputy Director of CIC Insights, analyzed for the Science and Technology Innovation Board Daily that Tencent's long-term strategy establishes a financial safety net for this massive investment, indicating that the computing power assets have liquidity and an exit strategy in the market. Purely renting computing power belongs to a low-barrier pipeline business, and the gross margin is greatly influenced by fluctuations in chip supply. Tencents choice to prioritize self-use is anticipated to yield commercial increment from utilizing computing power in its advantageous products, generating gross margins far exceeding rental income, Zhang Xiaolu stated. Emerging concerns: Short-term pain from heavy asset transformation On the other hand of clear opportunities lies the real pressure from the transformation into a heavy asset model, with short-term profitability being under significant strain. Financial data shows that Tencents GAAP net profit attributable to the parent grew only 0.7% year-on-year to 56.022 billion yuan in the second quarter, significantly lower than the 11% revenue growth, and falling short of market consensus expectations. Excluding the income, costs, and expenditures associated with new AI products such as Hunyuan, Yuanbao, WorkBuddy, CodeBuddy, and WeChat Xiaowei, Tencents Non-IFRS operating profit for the second quarter could have reached 86.1 billion yuan, a year-on-year increase of 19%. Calculating this way, the new AI business segment alone dragged down about 10.5 billion yuan in operating profit in the second quarter, equivalent to an average daily investment of over 100 million yuan. The longer-term challenge comes from depreciation pressures and the inherent switch in the business model. Historical data shows that prior to 2023, Tencent's annual capital expenditure consistently remained within the range of 20 billion to 30 billion yuan; it stood at only 23.9 billion yuan for the entire year of 2023; then soared to 76.76 billion yuan in 2024 and 79.198 billion yuan in 2025; by just the first half of 2026, it reached 84.72 billion yuan, likely to exceed 150 billion yuan for the entire year. In three years, the scale of capital expenditure has increased more than fivefold. Some institutional analysts are concerned that Tencent is passively transforming from a typical light-asset internet platform company into a heavy-asset AI infrastructure company under competitive pressure from the AI wave. As of the second quarter, the companys property, equipment, and facilities recorded a quarterly depreciation of 9.553 billion yuan, a sequential increase of 14.6%. At the current pace of investment, depreciation is likely to double in the next 2-3 years, creating long-term pressure on the profit and loss statement. Zhang Xiaolu pointed out to reporters from the Science and Technology Innovation Board Daily the timing mismatch between input and return: In the short term, standalone software cannot fully accommodate backend depreciation. The depreciation period for computing servers is 3-5 years, and even if WorkBuddy has the highest monthly active users, regarding the current B-end access situation in the domestic market, there is a significant difference in the scale of hardware depreciation. Market concerns also extend to the reconfiguration of valuation logic. Historically, Tencent, as a light-asset internet platform, has benefited from a long-term PE valuation central tendency of 20-25 times; if ongoing large-scale capital expenditure squeezes free cash flow and heavies asset attributes, the market may shift towards the pricing logic of infrastructure companies, applying downward pressure on the valuation center. The adjustment of Tencents stock price on August 13 following the semi-annual report reflects a concentrated release of such worries. Global computing power race: Market pricing logic shifts towards "return verification" The Science and Technology Innovation Board Daily also noted that Tencents hundred-billion betting is not an isolated case. Since the breakthrough of ChatGPT, a global computing power arms race among tech giants has begun, and the pricing benchmarks in the capital market are quietly changing. In the domestic market, Alibaba's cumulative capital expenditure for the fiscal year 2026 is approximately 120 billion yuan, having previously announced investments of 380 billion yuan over the next three years for cloud and AI infrastructure; ByteDance has raised its AI capital expenditure forecast for 2026 to over 200 billion yuan, an increase of more than 25% from its original plan. Billion-level investments have become standard for the leading domestic companies. Globally, the scale of investments by the four major North American tech giants is even more astounding: Microsoft, Amazon, Google, and Metas combined capital expenditures for 2026 approach 750 billion USD. Amazon's capital expenditure for the second quarter was 54.2 billion USD, a year-on-year increase of 69%, with its annual guidance raised to 220 billion USD; Microsofts second-quarter investment was 41 billion USD, a year-on-year increase of 70%, while signing over 130 billion USD in data center lease agreements during the same period. However, the feedback from the market has been markedly different: After the earnings reports, Amazon and Microsoft saw their stock prices surge by 15% and 9%, respectively; Google's second-quarter free cash flow turned negative for the first time since its IPO, with the stock price dropping over 4%; Meta's free cash flow plummeted by 91% year-on-year, resulting in over a 10% decline in after-hours trading. In simple terms, companies with verified AI demand enjoy valuation premiums, while those with high investment but unclear returns face corrections. For overseas tech giants, the contradiction in market revaluation is not about the authenticity of AI demand, but rather whether the rhythm of demand realization can outpace upfront capital investment, noted Wang Kai, Chief Analyst of Asset Allocation at EB SECURITIES, in an interview with the Science and Technology Innovation Board Daily. When capital expenditure for computing power infrastructure consumes free cash flow, against the backdrop of high US Treasury yields, the market transitions from listening to stories to watching capital return rates. For the application side, the real test lies in the unit economic model; those who can convert active users into essential subscriptions and transition enterprise procurement from pilot projects to large-scale rollout will capture the valuation premium. In the global landscape, Tencent finds itself in a middle ground: It has the WeChat ecosystem with 1.4 billion monthly active users as a landing scenario, WorkBuddy with preliminary validation of its commercialization capability, and a fallback option of renting out computing power at any time, yet the market has yet to see a clear return inflection point corresponding to the hundred billion investment. Investment bank perspectives are also diverging: Morgan Stanley expects flat earnings in the second half of 2026 to 2027, reducing its target price from 650 HKD to 550 HKD while maintaining an overweight rating; Goldman Sachs raised its capital expenditure forecasts for 2026-2028 to 210 billion, 245 billion, and 247 billion yuan, with its target price slightly decreasing from 700 HKD to 670 HKD; Citigroup, on the other hand, raised its target price to 765 HKD, believing that Tencent's AI plans are yielding significant results and its core business remains resilient. The next 6 to 12 months will be a critical verification period. The comprehensive rollout of WeChat Xiaowei, conversion data for WorkBuddys paid users, performance of Hy4, and the proportion of AI income from cloud services will collectively answer one question: Is the more than 80 billion yuan that Tencent is investing laying the foundation for a new moat for the next era, or is it another round of ineffective consumption in industry competition? The market will be watching closely.