Tencent President Martin Lau and other executives addressed the company's latest AI deployment plans and spending issues. On August 12, Tencent Holdings (00700.HK) reported its second-quarter earnings, with revenue reaching 204.79 billion yuan, an 11% year-on-year increase, and Non-IFRS operating profit hitting 75.64 billion yuan, up 9%. Excluding the revenue, costs, and expenses from new AI products (primarily Hy, Yuanbao, CodeBuddy, WorkBuddy, and Xiaowei), Non-IFRS operating profit grew 19% year-on-year to 86.1 billion yuan.

The market is focused on Tencent's aggressive AI spending: in the second quarter, the company's capital expenditure was 52.78 billion yuan, a 176% increase year-on-year and a 65% increase quarter-on-quarter. Free cash flow for the quarter was negative 13.8 billion yuan, but excluding prepayments for computing power procurement, it stood at 37.6 billion yuan. In the post-earnings call, Tencent's management, including Pony Ma and Martin Lau, collectively addressed market hot topics. As of the midday trading on August 13, TENCENT (00700.HK) shares fell 3.81% to close at 444 Hong Kong dollars.

Computing Power Can Be Resold for Profit, but Tencent Chooses Self-Use

Tencent's management stated that, driven by soaring demand for computing power and rising rental prices, if the company were to lease out its existing computing capacity, it could "almost immediately recover depreciation costs." They revealed that prepaid orders for computing power from several months ago could now be resold for a profit of over 30%, "higher than the original purchase price." However, Tencent has not opted for this short-term monetization path. The management noted that the priority for new computing power is: first, to train larger-scale Hunyuan models; second, to provide inference support for its own AI applications like WorkBuddy; and finally, to lease out any remaining capacity.

Martin Lau further explained the rationale behind this strategic choice: "By building the most advanced models and market-leading AI applications, we can deliver superior intelligence, which can ultimately be translated into long-term, substantial economic returns, such as through selling tokens in applications."

Hunyuan 4 Will Be Larger, Aiming to Gradually Approach SOTA

Regarding the model roadmap, Martin Lau explained that while Hunyuan 3 is considered a "very small model" by current standards, it possesses capabilities comparable to or even surpassing larger models. Its focus is on practical application scenarios rather than benchmark performance, "often proving more useful in real-world situations than many models of the same or even larger size." He stated that this principle will continue with Hunyuan 4, which will be a larger-scale model, not only capable of outperforming larger rivals but also surpassing Hunyuan 3 in practicality. A subsequent Hunyuan 5 will aim to gradually approach and ultimately achieve the state-of-the-art (SOTA) level. By then, Tencent will have a matrix of models of different sizes to solve various user problems at different cost levels, designed in coordination with products to improve execution efficiency.

In response to a Goldman Sachs analyst's question about differentiation at the 3-trillion-parameter level, Martin Lau did not disclose specific technical parameters, emphasizing that the Hunyuan series will continue to focus on practicality and product synergy as its core differentiation.

WorkBuddy Sees Explosive Growth, Investment in AI Products Increases to 10.5 Billion Yuan

In the second quarter, Tencent's investment in new AI products increased from approximately 8.8 billion yuan in the first quarter to about 10.5 billion yuan. Senior management revealed a "very significant change" in capital allocation: due to the explosive growth of the enterprise AI application WorkBuddy, Tencent decisively prioritized resources toward WorkBuddy while lowering the priority of other products in its AI portfolio. On profitability, management disclosed that the gross margin for WorkBuddy paid users is now comparable to the overall gross margin of Tencent Cloud. The overall gross margin is lower, mainly due to free users, as Tencent uses subsidies to gain market share and user growth.

Regarding WorkBuddy's positioning, Martin Lau stated that Tencent does not view it merely as daily office software but is continuously exploring high-value-added application scenarios, "even helping users make more money in some cases," to unlock more business models. Tencent's Chief Strategy Officer, James Mitchell, added that WorkBuddy's revenue is primarily subscription-based, similar to the gaming business, with a time lag between cash collection and revenue recognition. However, current cash revenue is growing significantly and is expected to be converted into Tencent Cloud's reported revenue by the end of the year. He also mentioned that by the end of this year and into next year, Tencent will have sufficient GPU and ASIC capacity, at which point it will increase its efforts in bare-metal GPU leasing and Model-as-a-Service (MaaS) businesses.

WeChat Xiaowei: An Ecosystem Amplifier in the AI Era

Martin Lau compared the impact of AI on WeChat to the value amplification Tencent's mobile internet era brought to QQ in the PC era. He pointed out that in the PC era, QQ was primarily a communication and social tool, while in the mobile era, WeChat amplified its value by more than ten times. He sees a similar opportunity for the WeChat ecosystem in the AI era, evolving from AI-driven to AI-first. "Currently, users need multiple clicks to complete an operation; in the future, they will only need to give a single command, and the system will execute the transaction automatically."

Lau described a future Agent-to-Agent scenario: users issue complex commands through the WeChat AI assistant "Xiaowei," while merchants and mini-programs have their own Agents. The Agents from both sides interact directly to complete transactions. In the longer term, every user will have their own dedicated Agent, and Agents can communicate with each other. Regarding on-device inference, Lau believes that shifting most inference to the device side will take a significant amount of time, but the trend is certain. As the GPU power of phones and computers increases, more inference will be done locally, and the industry will return to a "device-centric, cloud-assisted" norm, with computing power capital expenditure borne by the entire ecosystem rather than solely by model companies. He clarified that WeChat's self-developed WeLM model is not dependent on Hunyuan's SOTA capabilities but is designed around privacy protection, cost-effectiveness, and engineering needs within the WeChat ecosystem.

On the AI cloud business, Tencent's management stated that while domestic token prices are low, production costs are even lower, so the token business maintains a positive gross margin. They revealed that Tencent Cloud implemented a comprehensive price adjustment in May and significantly reduced discount levels, "The overall pricing environment in the domestic cloud market is no longer as difficult as it was in the past."

Dynamic Capital Allocation, No Profit Timeline Set

Regarding capital allocation, management responded that it will be dynamically adjusted. If AI capital expenditure can generate higher returns, the company will increase related investments and correspondingly reduce share buybacks. Martin Lau added that the capital expenditure for building the AI business is more concentrated in fixed-cost investments for this year and next, "It should not be assumed that there will be such a large new investment every year." Inference computing power investment is directly linked to business returns, and the scale will be adjusted if returns fall short of expectations. Funding sources will comprehensively consider operating cash flow, balance sheet cash, portfolio value, and prudent debt leverage levels.

When asked when AI investments will be reflected in profits, Lau did not provide a specific timeline. He stated that the cost of the WeChat AI assistant "Xiaowei" will be less than the investment in "Yuanbao" over the past year, and "the costs are completely controllable." He emphasized that Tencent will not provide specific performance guidance but will continue to control costs with its usual discipline. "If we were to simply change our business model today to purely leasing out computing power, we would not only avoid losses but would be profitable," Lau said. He added that leasing out computing power is always a backup option for Tencent, "This is our fallback at any time."