Tencent’s AI Spending Is Testing Investor Patience After Solid Q2 Results

Record AI spending turned cash flow negative

The most important warning signal was cash generation. Tencent reported capital expenditure of RMB52.8 billion for the quarter, while quarterly free cash flow fell to negative RMB13.8 billion as investment outpaced operating cash generation.

That does not prove the spending will destroy value. It does change the standard investors use to judge the strategy. Tencent now needs to demonstrate that AI can produce faster advertising growth, higher enterprise demand, better margins or new transaction revenue at a scale sufficient to justify continuing compute and model costs.

This is why Mizuho’s move is better understood as a valuation reset than as a simple downgrade to Tencent’s business outlook. One report said the brokerage reduced the valuation multiples applied to its 2027 estimates while leaving its fiscal-year earnings estimates largely unchanged.

HY3 provides evidence of adoption—but not yet monetisation

Tencent has made visible progress with its Hunyuan AI models. The production version of HY3 ranked among the top three models globally by token usage on OpenRouter, according to company and earnings-call materials. It is being used across products including WorkBuddy, Yuanbao, games and Weixin.

That is evidence of usage and distribution. It is not yet evidence of durable, high-margin revenue. Tencent describes HY3 as a cost-effective step toward future Hunyuan models with state-of-the-art capabilities, meaning the company is still investing toward its longer-term platform ambitions.

The next stage will require more than model popularity. Investors will want to see paid demand, improving unit economics and measurable benefits in Tencent’s existing businesses.

Xiaowei could unlock WeChat’s distribution—but remains experimental

The company’s most consequential AI option may be Xiaowei, an agentic assistant being tested on a small scale inside Weixin, the Chinese version of WeChat. Tencent has described potential use cases involving search, communications and mini-programs, but the product remains at the prototype-testing stage.

WeChat’s ecosystem could give a successful assistant an unusually broad route into search, commerce, advertising and services. For now, though, Xiaowei should be treated as future option value rather than a current earnings driver. There are not yet enough disclosed engagement, conversion or revenue figures to establish its commercial impact.

Gaming offers less of a cushion overseas

Tencent’s domestic gaming performance was strong, but international gaming revenue declined 0.8% year over year to RMB18.6 billion, although it grew 4% on a constant-currency basis.

The decline is relatively small, yet it matters in the context of AI spending. A weaker overseas gaming contribution reduces one source of diversification and makes investors more dependent on advertising, cloud and future AI monetisation to support the next investment cycle.

What investors are watching next

Tencent’s results have shifted the central debate. The question is no longer simply whether the company can build credible AI products. It is whether Tencent can convert model adoption and WeChat distribution into incremental revenue and margins quickly enough to offset the cost of developing larger models, expanding infrastructure and competing in fast-moving categories.

The bullish case has real support: revenue growth returned to double digits, advertising accelerated, domestic gaming remained strong and HY3 gained substantial usage. The cautious case is equally concrete: quarterly free cash flow turned negative, reported profit growth was modest, international gaming weakened and key products such as Xiaowei remain experimental.

That balance explains Mizuho’s Neutral rating and lower HK$560 target. Tencent’s AI strategy may eventually strengthen the company’s ecosystem, but investors are being asked to fund a capital-heavy transition before the returns are clearly visible.