Why Chinese AI labs bet on open weights
Unlike closed frontier labs in the West, Chinese labs use open weights as their primary distribution strategy to build developer mindshare and win enterprise workloads.
The open-weight consensus
While leading US labs keep frontier model weights closed behind paid APIs, nearly every major Chinese AI organization releases open weights: Alibaba's Qwen family, DeepSeek, Zhipu GLM, 01.AI's Yi, and ModelBest's MiniCPM.
Releasing weights under permissive licenses (such as Apache 2.0 or MIT) allows developers worldwide to inspect architectures, fine-tune on custom data, and deploy locally without API vendor lock-in.
The business strategy
Open weights solve the distribution challenge for Chinese labs. Western enterprises often cannot send private data to overseas APIs, but they readily download open weights for self-hosting. This has made Qwen and DeepSeek default open-source baselines globally.
For cloud providers like Alibaba Cloud and Volcano Engine, open weights drive massive cloud consumption: developers fine-tune and run inference on their infrastructure, monetizing compute rather than raw API access.