What Ma Huateng (马化腾) Said About "Shrimp Farming" Is More Than Just Shrimp Farming

Financial backdrop: Tencent's steady fortress

Tencent (腾讯) reported a strong 2025: revenue of ¥7517.66 billion, up 13.86%, and attributable net profit of ¥2248.42 billion, up 15.85%. Games, advertising and fintech remain the cash engines — domestic game revenue rose 18%, advertising 19%, and fintech and enterprise services grew 8%. In Western shorthand: a trillion‑plus valuation company that still looks like the mobile‑internet winner it has been for a decade. But Ma Huateng (马化腾) used a surprising folk metaphor — “养虾” or “raising shrimp/ lobsters” — at the earnings call to flag a deeper strategic shift.

The metaphor and the move to distributed Agents

Ma framed the “lobster” phenomenon not as a single product test but as a mirror for a larger architectural change: moving from a single‑entry super‑app model to a distributed Agent ecosystem embedded across WeChat, QQ, Enterprise WeChat and other Tencent properties. It has been reported that an early AI app — referenced as 元宝 — briefly climbed to top AI app daily active user rankings during Lunar New Year but then lost steam, illustrating how a hit‑and‑run pattern differs from durable ecosystem integration. Ma’s point: every mini‑app or small program (小程序) can be “智能化/龙虾化” — each product grows its own Agent tailored to its scene rather than funneling everything through one unified AI entrance.

Strategy, risk and the middlemen problem

Tencent is reportedly investing across three layers: the model base (its Mengtian/mixed model stack), scenario‑level Agents embedded in specific apps (meeting notes, document drafting, enterprise customer service), and an open developer layer that lets third parties build Agents inside Tencent’s soil — a repeat of the small‑programs playbook. But there is a tradeoff. If Agents become smart enough to bypass intermediaries, service providers, content creators and mini‑app operators risk being “short‑circuited.” Who finishes the task — the system or the original service provider — is becoming the decisive competitive axis. Can Tencent preserve partners’ value while letting AI automate tasks? Ma’s answer: a calibrated mix of decentralization and selective centralization.

Geopolitics, models and why Tencent must do it itself

This is happening against a broader tech rivalry and export‑control backdrop: US restrictions on advanced chips and model components make reliance on external model providers risky, Ma warned implicitly — without its own model stack, Tencent would be “renting someone else’s land.” So the company is choosing a slower, messier path: upgrade its vast mobile‑era ecosystem into a distributed Agent network rather than chase a centralized Nvidia/OpenAI‑style stack. The result will shape not only Tencent’s future but the fortunes of the millions who build on its platform. Who controls completion of the task — not just access to users — will determine who wins in China’s next internet era.