Tencent has rebuilt its AI model infrastructure over six months, prioritizing data quality and cost-effective performance. The first major output is the Hunyuan Hy3 Preview, a practical, open-source mixture-of-experts model designed for deployment across Tencent's products like QQ and Tencent News. The company is accelerating the release of AI agents, including WorkBuddy, QClaw, and CodeBuddy, and is also developing world models and embodied AI. The core business challenge is that AI has high variable costs, making free distribution unsustainable. Tencent's strategy rests on three pillars: anchoring on high-value, scalable use cases like work productivity; balancing aggressive market share growth with incremental revenue capture; and preparing for large, ongoing investments in a competitive arena where costs will escalate. The speaker notes that Tencent's existing profitable businesses provide a funding advantage over competitors like OpenAI, which rely on external fundraising.