Tencent’s Yao Shunyu: AI Moats Are Built in Products, Not Models
TREE NEWS reports: Tencent Chief AI Scientist Yao Shunyu argued that the real competitive moat in artificial intelligence lies at the product layer rather than the model layer, using WeChat as a case study. Speaking at a dialogue with Zhang Ya-Qin during the 2026 Future Science Prize events, Yao said that an AI embedded in WeChat does not necessarily need the strongest underlying model; what matters is its access to user schedules, social graphs, and other proprietary context that a competitor with a superior model cannot easily replicate.
Yao, who now oversees Tencent’s foundation model division, added that even a rival launching a more powerful model would struggle to displace an incumbent that owns the user relationship and data flywheel. Zhang responded that, by that logic, Anthropic might not be worth the $1 trillion to $2 trillion valuations sometimes floated for it.
Why the Product-Layer Thesis Matters
The exchange cuts to a central debate in the AI industry: whether value accrues to model developers or to the applications that own distribution and context. Yao’s view implies that model capabilities are commoditizing faster than distribution, and that proprietary user data and workflow integration create stickier advantages than benchmark scores.
- Distribution beats raw capability: Products with existing user bases can deploy “good enough” models and still win on context and convenience.
- Data flywheels compound: Social graphs, calendars, and transaction histories are hard to copy and improve personalization over time.
- Valuation risk for pure model labs: If moats sit in products, standalone model companies face pressure to move up the stack or justify premium multiples.
Implications for Crypto AI
The thesis has direct relevance to crypto’s AI sector, where decentralized compute networks, inference marketplaces, and on-chain agent protocols often pitch tokenized models or GPU capacity as the core asset. If Yao is right, those projects may need to capture the application layer — wallets, agents, and consumer interfaces — rather than compete purely on model quality or compute pricing.
Decentralized AI networks that settle payments on-chain could still thrive as infrastructure, but the durable margin may accrue to whoever owns the end-user relationship. Crypto projects that tokenize inference or model access without a distribution strategy risk being squeezed as open and proprietary models alike become cheaper and more capable.
Forward Look
Expect more AI labs to pivot toward product integration and enterprise workflows, while crypto AI teams emphasize agent frameworks and consumer-facing applications over raw model benchmarks. The debate over where AI value accrues — and how it is captured on-chain — is only beginning.
