China’s open-weights AI strategy is winning

China is taking a radically different approach to AI development than the United States, and it might be working. While American companies like OpenAI and Anthropic guard their AI models behind paywalls and APIs, Chinese firms are releasing open-weights models that anyone can download, modify, and run locally.

What Open-Weights Actually Means

Open-weights models differ from traditional open-source software. Companies release the trained neural network weights - the mathematical parameters that make the AI work - but not necessarily the training code or datasets. This means developers can use and modify the models without paying per API call or sending data to external servers.

Meta's Llama series pioneered this approach in the West, but Chinese companies have embraced it more systematically. Models like Qwen from Alibaba and ChatGLM from Zhipu AI are freely available for download. Developers can run them on their own hardware or through dozens of hosting providers competing on price and features.

The Economics Behind the Strategy

This strategy flips the traditional software business model. Instead of charging for access to the AI, Chinese companies make money from cloud hosting, enterprise support, and specialized applications built on top of the base models. The approach creates a massive ecosystem where hundreds of providers can offer the same underlying AI capabilities.

The economics work because hosting and fine-tuning services become commoditized. When multiple companies can offer the same model, they compete on service quality and price rather than model performance. This drives down costs for end users while creating more opportunities for smaller players to enter the market.

Chinese firms also benefit from domestic data and regulatory advantages. They can train on Chinese internet content and serve Chinese customers without the compliance overhead that American companies face when operating in China.

Pressure on Proprietary Systems

The open-weights strategy puts direct pressure on companies that rely on API-only access models. When developers can get comparable AI capabilities for free, paying monthly subscriptions for ChatGPT Plus or Claude Pro becomes harder to justify for many use cases.

Enterprise customers particularly benefit from open-weights models. They can ensure data never leaves their infrastructure, customize models for specific domains, and avoid vendor lock-in. These advantages matter more to business customers than the marginal performance differences between competing models.

The approach also accelerates innovation cycles. When researchers worldwide can experiment with and improve the same base models, capabilities advance faster than when development happens behind closed doors at individual companies.

Werd.io argues this strategy follows historical patterns where open systems eventually overtake proprietary ones. The shift makes AI development more accessible to smaller companies and researchers while forcing proprietary providers to justify their premium pricing through superior performance or unique capabilities.