Leaking Ship, Still Aboard: Ma Huateng’s Candid Take on Tencent’s AI Turnaround

On May 13, during Tencent's shareholder meeting, Ma Huateng faced sharp investor questions about whether Tencent AI was falling behind. He answered with a candid metaphor about a leaking ship, signaling a fundamental shift in Tencent’s attitude toward AI.

Leaking Ship: From "Thinking We’re Onboard" to "Discovering Leaks"

He described three stages: In 2024 the release of the Mixed‑yuan large model gave the impression of being on the AI wave, but the word "thinking" revealed it was merely a pose on the dock. Underlying capabilities were weak, compute investment conservative, data silos and cross‑department friction caused systemic leaks. Now the technology works but commercial loops and user experience remain unstable. Data shows the AI assistant Yuanbao has 57.35 million MAU, while ByteDance’s Doubao exceeds 100 million; a 10 billion‑yuan red‑packet campaign yielded only 18.7 % 30‑day retention.

Repairing the Ship: 36 Billion and One Person

At the end of 2025 Tencent hired former OpenAI researcher Yao Shunyu as chief AI scientist reporting directly to President Liu Chifeng, elevating AI from a division‑level to a company‑wide strategy. The company created new AI Infra and AI Data divisions, dissolved the AI Lab, and merged all teams into the Mixed‑yuan group, ending internal competition. Capital investment rose to 180 billion yuan in 2025 for Mixed‑yuan and Yuanbao, projected to double to 360 billion yuan in 2026. Q1 2025 capex reached 319.4 billion yuan, with AI‑related capex at 370 billion yuan and R&D spending 225.4 billion yuan (+19 % YoY).

Yao emphasized “reducing leaderboard chasing, focusing on stable real‑world scenarios.” The April preview of Mixed‑yuan Hy3 (MoE architecture, 295 billion total parameters, 210 billion active parameters, 256 K context length) achieved a ten‑fold increase in token calls versus the previous generation and a 16.5× rise in code and agent usage, topping OpenRouter with 3.66 trillion tokens and winning programming and tool‑call categories.

Boarding: The "Parasitic" Strategy in the WeChat Ecosystem

Instead of building a new platform, Tencent embedded AI agents in its super‑carrier WeChat. QClaw, a consumer‑facing AI agent, reached one million domestic users in ten days and five million overseas in two weeks, operating via QR‑code binding and chat‑window control without a separate app. WorkBuddy, aimed at workplace productivity, scaled compute ten‑fold on launch and became China’s most popular AI efficiency service. Yuanbao added a chat‑record summarization feature inside WeChat, leveraging the 1.418 billion‑monthly‑active‑user ecosystem as a natural entry barrier.

This parasitic approach mirrors earlier strategies (mini‑programs, video channels) and reflects Ma’s comment that Tencent will rely on its 1.4 billion‑user WeChat moat rather than chase external territory.

Still Not Sitting Down: Commercialization Reefs

Despite technical breakthroughs, AI revenue dilutes overall profit: Non‑IFRS operating profit was 756.3 billion yuan (+9 % YoY); removing AI income, costs, and expenses yields 844 billion yuan (+17 % YoY). The 360 billion yuan AI spend accounts for an 8‑percentage‑point profit gap. Analysts at Goldman Sachs and Bloomberg cut 2026 profit‑growth forecasts, citing unclear monetization paths. While the WeChat ecosystem is a moat, it may become a ceiling as AI functions are limited to plugins, restricting independent app innovation.

Can the Ship Speed Up?

Ma concluded with hope for a faster ship, leaving open questions about whether the 360 billion yuan investment will accelerate progress, whether the WeChat‑centric parasitic model can achieve a full commercial loop, and whether Yao’s stability‑first roadmap can keep Tencent competitive without leaderboard chasing.