Recently, Tencent Holdings released its financial report for the second quarter of 2026, delivering an impressive report card. The company achieved revenue of 204.79 billion yuan for the quarter, an 11% year-on-year increase, with quarterly revenue surpassing 200 billion yuan for the first time; Non-IFRS operating profit was 75.64 billion yuan, up 9% year-on-year.

However, compared to its impressive revenue figures, Tencent's "spending money" in AI has attracted even more market attention. The financial report shows that Tencent's capital expenditure in the second quarter reached 52.78 billion yuan, a year-on-year surge of 176%; R&D expenditure also reached 27.28 billion yuan, a year-on-year increase of 35%.

Facing questions about "when the massive AI investment will break even," Tencent President Martin Lau did not provide a precise timeline during the earnings call.

| Financial Report "Split in Two": Main Business Steady, AI Investment Drags Down Profits in the Short Term

Tencent's Q2 financial report structure underwent a significant change, clearly splitting it into two layers: "mature core business" and "new AI business."

In its mature core business, which has been active for over a decade, performance remains steady: domestic game revenue grew 17% year-on-year to 47.3 billion yuan, with daily active users for "Delta Force: Hawk Ops" and "Valorant" hitting record highs; Marketing services revenue increased 22% year-on-year to 43.57 billion yuan; fintech and enterprise services revenue increased 9% year-on-year to 60.29 billion yuan.

However, Tencent's massive investment in AI has indeed caused significant disruptions to short-term profits. The financial report candidly disclosed that new AI products, including Hunyuan, Yuanbao, and CodeBuddy, impacted Non-IFRS operating profit by 10.5 billion yuan this quarter. Excluding the revenue, costs, and expenses of these AI products, Tencent's mature core business—Non-IFRS operating profit—actually grew 19% year-on-year, reaching 86.1 billion yuan.

To support AI development, Tencent's investment in infrastructure is rapidly increasing. Based on single-quarter scale, its annualized computing power procurement has exceeded 200 billion yuan. Management clearly stated that this is a "phased concentrated investment" in large model training and inference computing power reserves. This high capital expenditure directly led to profit fluctuations in the financial reports and put pressure on Tencent's free cash flow in the short term.

In response to analysts' attention to the massive investment, Tencent's management provided clear logic and two sets of risk hedging monetization plans. First is "priority self-use," supplying computing power to self-developed large models and proprietary products like WorkBuddy, earning long-term subscription and token income by building a platform ecosystem. Next is the "guaranteed minimum plan," which means leasing idle computing power externally.

Management revealed that a batch of computing power equipment ordered with a deposit a few months ago is now being resold for over 30% profit compared to the original purchase price. This means that even if the development of proprietary applications falls short of expectations, computing power leasing itself has a solid foundation for profitability.

| Model and Agent Dual Racing Line: Hunyuan Iteration Accelerates

While investing heavily to stockpile computing power, Tencent has also launched a "three-pronged" surge in AI model and application layers.

At the model layer, after rebuilding its technical foundation, the Hunyuan large model underwent rapid iteration. Its latest Hy3 version completed the upgrade from preview to official in less than three months, not only competing with flagship models with 2 to 5 times the parameter count on the market with a relatively small scale but also increasing daily usage by sevenfold, firmly ranking among the top three globally in token consumption.

Martin Lau also revealed that the Hy4 version, with stronger performance and a fundamental change in practicality, will be launched soon. More importantly, Tencent has 'co-designed' models and products. For example, after Hy3 was integrated into the office agent WorkBuddy, its task resolution rate soared from 72% to 90%, making it the platform's preferred model.

At the application layer, Tencent's AI office agent has already demonstrated a clear leading edge. According to industry reports, WorkBuddy's PC visits in June exceeded 20 million, ranking first among similar domestic products and even surpassing the combined total of the second and third place competitors.

Currently, Tencent has built a powerful office intelligent suite matrix, with WorkBuddy, CodeBuddy, QClaw, and Marvis firmly occupying four spots in the top ten monthly traffic charts, forming an overwhelming scale advantage in the AI office sector.

| Market Divergence: Is It the 'Eve of Harvest' or a 'Value Trap'?

Tencent's massive investment in AI has sparked the most fundamental concern in the capital market: when will such aggressive investment yield tangible returns?

After the financial report was released, Tencent's stock price opened lower and continued to fall on the first trading day after the Hong Kong stock market earnings. The logic of the market "voting with its feet" is very straightforward: the new AI business in Q2 directly dragged down about 10.5 billion yuan in operating profit, and including computing power prepayments, free cash flow even dropped to -13.8 billion yuan. Although the core business still generates 37.6 billion yuan in positive cash flow after excluding prepayments, the heavy asset management of AI has indeed squeezed shareholder returns in the short term.

Martin Lau also directly addressed the market's concerns. He made it clear that large-scale concentrated AI investment will only last for this year and next, and will not remain the same scale for long; subsequent increases in inference computing power will depend entirely on actual business returns.

Kingtech Perspective | Focus on "Guaranteed Plans" and "Agent Monetization"

Tencent's management proposed a dual-track system of "minimum deposit for computing power leasing" and "priority for self-use," providing a safety cushion for this massive investment.

In terms of investment, investors are advised to closely monitor two validator nodes:

First, the actual revenue conversion from Tencent Cloud's computing power leasing business. If idle computing power can be quickly monetized, it will greatly ease market cash flow concerns.

Second, the payment conversion rate of AI office agents (such as WorkBuddy). As Tencent's AI application layer currently has the strongest scale advantage, the speed of commercialization will directly determine whether Tencent's AI business can move from a "cost center" to a "profit center."

Currently, market concerns about Tencent AI mainly focus on short-term EPS (earnings per share) pressure, and some investment banks (such as Morgan Stanley) have lowered their target prices. But as Citigroup and other institutions have pointed out, Tencent's core fundamentals remain resilient, and its AI strategy has already shown results on the application side. At current valuation levels, the market barely reflects AI's future value.

For long-term funds, this "expectation gap" caused by short-term earnings pains actually offers pricing opportunities to "trade time for space." It is recommended to buy on dips and gradually position when stock prices pull back due to pressure from AI investment, patiently waiting for the overall AI business break-even in 2027.