Chinese open AI models eclipse US rivals in enterprise use

Chinese open-weight AI models now account for 41% of downloads on the Hugging Face platform, overtaking US models this spring. On the OpenRouter aggregator, the six most popular models are open-weight systems from Chinese firms like Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai, leaving Anthropic’s Claude Opus 4.7 in seventh place. The data indicates a decisive shift in where developers are actually deploying AI.

For businesses, this represents a potential escape from expensive vendor lock-in with American tech giants. Open models handled nearly a third of AI requests on the Vercel platform in June, taking over high-volume infrastructure tasks while closed models are relegated to a premium, higher-cost tier. Half of the Fortune 500 are already using Hugging Face to deploy private or open-source models instead of renting access.

Cost and control are driving the transition. Clem Delangue, CEO of Hugging Face, noted that companies are rethinking their strategies after receiving the bills for scaling closed frontier models. “If you’re an AI company or a technology company, you don’t want to outsource your core capabilities to another company, to a black box API that you don’t control, don’t have any visibility on, and don’t really have any sort of ownership,” Delangue said.

The economics of open alternatives

The trend is accelerating as Chinese labs release increasingly capable models that are cheaper to customize than proprietary alternatives. Beijing-based Z.ai recently released an open-weight model called GLM-5.2 that competes with Anthropic’s latest releases on identifying security vulnerabilities and excels at agentic coding. This undercuts the economic advantage of the proprietary AI systems that US firms have spent billions developing.

The push against single-provider dependence is gaining traction among major tech executives. Microsoft CEO Satya Nadella warned that allowing learning to flow in only one direction means economic value converges toward the infrastructure owners rather than the creators of knowledge. “Therefore, it’s imperative that we distribute the learning infrastructure to every firm so that they can control their own learning loop,” Nadella said.

The safety dilemma

The rise of open models has triggered a fierce debate over AI safety. Anthropic CEO Dario Amodei has warned that releasing powerful open weights is dangerous because the models become difficult to control and accessible to bad actors for cyber or biological warfare.

Delangue rejects the idea that keeping models closed reduces risk. “The biggest risk in AI is concentration of power,” he said, arguing that transparency allows defenders to patch known cybersecurity exploits. “You don’t really make it safe by keeping it behind closed doors for just a few players. You make it more dangerous because you create asymmetry of power and asymmetry of capabilities.”