---
title: "Tencent's AI push lifts H1 revenue past ¥364B as capex triples"
date: 2026-10-27
category: Companies & Stack
site: NeuroAI
canonical: https://neuroai.site/a/na-tencent-h1-2025-ai-capex-payoff
language: en
---

# Tencent's AI push lifts H1 revenue past ¥364B as capex triples

> Tencent reported H1 2025 revenue of 364.5 billion yuan (≈US$50.3B) with net profit up 16%, even as AI capital expenditure more than doubled.

By the middle of August 2025, Tencent's finance team had a story that contradicted the usual AI narrative: heavy spending and fatter margins at the same time.

Half-year revenue reached 364.5 billion yuan (≈US$50.3B / HK$392B), up 14%, while non-IFRS net profit rose 16% to 124.4 billion yuan (≈US$17.2B).

Behind the numbers sat a capital-expenditure surge — second-quarter capex jumped 119% to 19.1 billion yuan (≈US$2.6B) — that President Martin Lau (刘炽平) insists is being spent "smartly," not blindly.

## Key takeaways

- Tencent's H1 2025 revenue rose 14% to 364.5 billion yuan (≈US$50.3B); non-IFRS net profit climbed 16% to 124.4 billion yuan (≈US$17.2B).

- AI-linked capital expenditure more than doubled: Q2 capex hit 19.1 billion yuan (≈US$2.6B), up 119% year-on-year, with H1 capex at 46.6 billion yuan.

- Margins improved even while spending — Q2 gross margin reached 57%, up from 53% a year earlier.

- Chairman and CEO Pony Ma (马化腾) credited AI with lifting the core businesses, and the group is folding models into WeChat, games, and ads.

## The earnings that surprised the skeptics

Tencent reported second-quarter revenue of 184.5 billion yuan (≈US$25.5B), up 15%, with non-IFRS operating profit also growing in the high teens. The detail that caught analysts was the margin direction. Gross margin expanded to 57% in Q2 from 53% a year earlier, meaning the business got more profitable even as it absorbed a much larger compute bill. Management attributed the improvement to higher-margin revenue mix and better cost efficiency in payments and cloud.

## Where the AI money went

The spending shows up in two lines. Research and development reached 20.3 billion yuan (≈US$2.8B) in Q2, up 17%. Capital expenditure — data centers, servers, land, and related infrastructure — totaled 19.1 billion yuan in Q2 and 46.6 billion yuan for the half. Cumulatively, since the group accelerated its AI strategy in the fourth quarter of 2024, AI-oriented capex had reached roughly 83 billion yuan by mid-2025.

President Martin Lau addressed the obvious worry head-on on the earnings call: Tencent must "spend smart," he said, rather than simply burn cash on chips, headcount, or marketing. The phrasing matters because it signals discipline, not an open chequebook.

## AI inside the core businesses

The payoff narrative rests on three segments. Value-added services — dominated by games — grew 16% in Q2 to 91.4 billion yuan, with game revenue up 22%, its third straight quarter above 20%. International games jumped 35% to 18.8 billion yuan, a fourth consecutive record. Marketing services rose 20% to 35.8 billion yuan, credited to AI-driven ad targeting across WeChat's video accounts, mini-programs, and search. Fintech and business services grew 10% to 55.5 billion yuan, helped by enterprise demand for GPU rental and API tokens.

WeChat and WeChat combined reached 1.411 billion monthly active users at the end of Q2. That installed base is the distribution rail for Tencent's consumer AI: the Yuanbao (元宝) assistant, AI search inside WeChat, and image-generation tools for official accounts.

Seen against peers, Tencent's path is distinct. Alibaba is spending upfront on infrastructure; ByteDance is pouring users into a consumer app. Tencent is moving more incrementally: embedding models into products it already owns, so the AI line item shows up as higher engagement and better ad yield rather than as a separate product bet. That is precisely why management can claim the spend is already paying back, instead of defending it as a long-term option. It is also a lower-risk posture — the group is not betting the quarter on a brand-new platform, but on making its existing ones smarter.

## The Yuanbao (元宝) and Hunyuan (混元) question

Tencent's model story splits in two. Hunyuan (混元) is the in-house large model (大模型) family, spanning fast "turbo" variants and deeper "thinking" variants, used to power internal products. Yuanbao is the consumer-facing assistant app, which saw a user surge after heavy promotion in early 2025 and then shifted toward retention and product polish in Q2.

The group's bet is that owning the model, the app, and the social distribution together is defensible in a way that a standalone chatbot is not. Whether Yuanbao converts free users into paying ones remains the open question, and management has acknowledged that Western-style per-user AI subscriptions are hard to sell in China.

## Honest limitations

Figures here come from Tencent's H1 and Q2 2025 earnings release and subsequent management commentary carried by Reuters and major Chinese financial media. "Non-IFRS" profit excludes certain items and is not a strict GAAP equivalent. The roughly 83 billion yuan cumulative AI capex figure is reported by Chinese media based on the earnings call and may include broader infrastructure, not only AI-specific spend. We have not independently audited the segment allocation. User and engagement numbers are company-reported.

## What readers can do now

- Watch Tencent's future quarterly capex and gross-margin pairings — if capex keeps climbing while margin holds, the "spend smart" claim is being validated in the numbers.

- Try Yuanbao and WeChat's built-in AI features directly; Tencent is shipping model capability into an app billions already open daily, which is the real distribution test.

- Enterprise buyers should track Tencent Cloud's GPU-rental and API-token pricing as a second benchmark alongside Alibaba's, since competition between the two should keep inference costs falling.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-tencent-h1-2025-ai-capex-payoff
Free to quote with attribution and a link to the original.
