China's "AI Made in China" Strategy: Openness as a Weapon

Beijing is using open-weight models to undermine American AI dominance. It is working better than expected.

Published: 25 July 2026 Category: AI Policy / Geopolitics Sources: Seoul Economic Daily

The Declaration

Chinese President Xi Jinping declared "AI Made in China" as a national priority this month, and the practical expression of that priority is pouring out of Chinese labs at a remarkable pace. Kimi K3. Qwen 3.8. GLM 5.2. Dozens of smaller models from labs that did not exist two years ago. All openly licensed. All globally available. All competitive with American frontier models on at least some dimensions.

The strategy is not subtle. Where the US has focused on building the most capable models and controlling access to them, China has focused on building capable-enough models and making them available to everyone. The US offers superiority. China offers access.

The Mechanics

Open-weight models are particularly suited to China's strategy because they bypass American control points. API-based models can be restricted through export controls, cloud provider terms of service, and IP geoblocking. Downloadable weights cannot. Once a model is on the internet, it is everywhere, and American regulators have no mechanism to prevent a developer in Kenya from downloading Qwen 3.8 and fine-tuning it for local use.

This creates a two-tier global AI landscape. Developers in US-aligned countries use GPT-5.6, Claude Fable 5, and Gemini 3.5 through APIs, with all the pricing, monitoring, and restrictions that entails. Developers everywhere else use Chinese open-weight models locally, with no API costs, no usage tracking, and no American oversight.

The Analysis

The US response has been confused. Export controls on chips have slowed Chinese training but not stopped it — Chinese labs have adapted through efficiency improvements, chip recycling, and domestic manufacturing. Attempts to regulate open-weight models have foundered on First Amendment and practical enforcement problems. The US has the better models but lacks a strategy for making them globally dominant.

China's approach has risks. Open-weight models can be fine-tuned for harmful purposes. They can be stripped of safety guardrails. They can be used to build systems that compete with Chinese state interests. Beijing is betting that the benefits of global distribution — influence, adoption, ecosystem development — outweigh these risks. So far, the bet is paying off.

The Verdict

The "AI Made in China" strategy is not about building the best model. It is about building the most distributed model. The distinction matters because the metrics of AI competition are changing. Peak capability still matters for prestige and certain applications. But global adoption, ecosystem development, and standard-setting matter for long-term influence, and China is winning on those metrics.

The US can respond by making its own models more accessible. It has not, partly because of safety concerns, partly because of commercial interests, and partly because of institutional inertia. The result is a world where Chinese open-weight models are becoming the default for global developers, not because they are better, but because they are available.

Openness is a competitive weapon. China is wielding it skillfully. The US has not yet figured out how to respond.

DEMYSTIFY