韩国媒体The Elec发文回顾腾讯Q2财报：游戏收入投入AI，CAPEX同比增176%

China's Tencent is pouring earnings from its gaming business into artificial intelligence (AI) investment.
According to Tencent's recently announced second-quarter results, domestic gaming revenue reached 9.7 trillion won, up 17% from a year earlier. International gaming revenue rose 4% to 3.8 trillion won. Combined gaming revenue totaled 13.5 trillion won. Growth in China was driven by titles including Delta Force, VALORANT on PC and mobile, and Roco Kingdom: World.
Tencent's AI investment is growing at an even faster pace. Capital expenditure (CAPEX) reached about 11 trillion won in the second quarter, up 176% from a year earlier. Combined with 6.57 trillion won in the first quarter, first-half CAPEX totaled 17.44 trillion won, surpassing the 16.31 trillion won spent in all of last year.
While not all CAPEX can be classified as AI investment, Tencent said securing AI computing capacity was a key driver of the increase. The company said it made substantial prepayments for AI-related computing to advance its internally developed Hy3 AI model and meet demand from AI productivity tools, Weixin AI services, AI deployment across its own products and external cloud customers.
The investment burden was also reflected in cash flow. Tencent recorded a net free cash flow (FCF) outflow of 2.84 trillion won in the second quarter. Excluding prepayments for computing purchases, however, FCF would have been positive at 7.74 trillion won, according to the company.
Gaming Revenue → AI Investment → In-House AI → Gaming Applications → Gaming Growth
Notably, the AI capabilities built through those investments are being fed back into Tencent's gaming business.
Tencent's latest internally developed foundation AI model is Hy3. The company reorganized its existing Hunyuan branding under the Hy name and officially released the model in July. Hy3 uses a mixture-of-experts (MoE) architecture, activating 21 billion of its 295 billion total parameters and supporting a context window of up to 256K. Rather than developing an AI model first and finding applications later, Tencent built Hy3 from the outset around functions needed across its own services, including Weixin, gaming and cloud.
Tencent executives said during the second-quarter earnings call that the company is "using Hy3 to create AI teammates in games and for code reviews," citing Peacekeeper Elite as a representative use case. The application goes beyond coding tools that assist developers. Tencent is deploying its own foundation model to create AI-controlled players inside games. In effect, AI users powered by Tencent's own model operate within its games.
Tencent also uses AI to generate 3D assets from text, images and sketches. The company said the technology can reduce 3D production work that previously took days or weeks to just minutes. The model is also offered to external developers through Tencent Cloud. It can generate cities and scenes, 2D images, animations, stories and dialogue, while supporting game testing and gameplay simulations. Players can also use Tencent's AI models to create user-generated content (UGC).
Tencent's AI strategy therefore differs structurally from that of rival game companies that purchase external AI models to improve development productivity. The cycle runs from gaming to AI investment, in-house AI development and back to gaming. Existing businesses, including games, generate cash to fund AI infrastructure and model development, while the resulting AI is deployed for game code reviews, 3D production, character animation, AI teammates and game assistants.
Tencent Cloud is another pillar of this cycle. Tencent's fintech and business services revenue reached 12 trillion won in the second quarter, up 9% from a year earlier. The company attributed the growth partly to higher cloud revenue driven by demand for AI-related services. Computing infrastructure built for its own AI development is also generating cloud revenue from external enterprise customers.
Tencent plans to continue investing in AI for the time being, though the company said spending at the current scale is not expected to recur every year.
Tencent President Martin Lau said during the second-quarter earnings call on Aug. 12 that "CAPEX for building AI-native businesses represents an initial batch of investment concentrated this year and next year." The company needs to secure computing resources for model training upfront, while subsequent investment in areas such as inference computing will depend on the returns generated by actual businesses.
Tencent is also not limiting its potential investment recovery to its own AI services. The company said it will prioritize its computing resources for training proprietary AI models and running AI applications, but can make the capacity available externally through Tencent Cloud when needed.
Tencent Chief Strategy Officer (CSO) James Mitchell said that, given recent rental prices for AI computing, the company could quickly recover depreciation costs even by renting capacity to external customers. He added that the market price of some computing resources acquired several months ago has risen more than 30% from their purchase price.
Overseas analysts see whether Tencent's existing businesses can sustain its large-scale AI investment as a key variable. Singapore-based financial group DBS said in an Aug. 13 report that 17% growth in domestic gaming and 22% growth in advertising showed that Tencent's core businesses were supporting its AI investment. DBS maintained its "Buy" rating but lowered its price target to HK$690 from HK$770, reflecting near-term cost pressure from AI investment.
Morgan Stanley said in an Aug. 13 report that Tencent is expected to expand AI spending, including AI model development, in the second half. It estimated CAPEX of 41 trillion won this year and another 41 trillion won next year, for a combined 82 trillion won. Morgan Stanley expects the AI investment burden to limit earnings growth from the second half of this year through 2027 and lowered its price target to HK$550 from HK$650.