Chinese AI companies are significantly shifting the global landscape by expanding their developer ecosystems through the promotion of open-source models, according to a report by Hyunwoo Choo from DigitalToday. While U.S. firms adhere to a closed API-based strategy, companies in China are making strides by offering models with downloadable open-source code that can be run and modified on users’ own hardware. This trend is drawing increasing attention as developers seek more customizable and cost-effective AI solutions.
The turning point in this landscape came with the release of the R1 reasoning model by DeepSeek in January 2025. The R1 model has been recognized for its performance that rivals leading U.S. models while operating at a lower cost, thereby narrowing the performance gap between Chinese and U.S. AI laboratories. This move not only enhanced developer interest but also set a precedent that other Chinese companies have followed.
Subsequent to DeepSeek’s success, several firms, including Z.ai, Moonshot, Alibaba’s Qwen, and MiniMax, have adopted similar strategies to produce higher-performance AI models aimed at competing directly with their U.S. counterparts. As the focus of the AI industry transitions from pilot projects to actual deployment and integration, the advantages of open-source models—such as lower costs and ease of customization—are becoming increasingly pronounced.
Quantitative indicators reflect this shift. Researchers from MIT and Hugging Face report that Chinese open-weight models accounted for 17.1% of global AI model downloads in the year leading up to August 2025. This figure has surpassed the U.S. share of 15.86%, marking a significant milestone where China has taken the lead in this particular metric for the first time. Data from Hugging Face also indicates that the number of user-generated derivative models based on Alibaba and the Qwen family has exceeded the combined total for models developed by Google and Meta.
Despite these gains, challenges remain. The development of Chinese models is often constrained by the country’s content censorship regulations, which influence the training data and outputs to align with governmental policy. In February, Anthropic alleged that some Chinese labs had improperly extracted performance benchmarks from its Claude model using a technique known as distillation, where one model’s outputs are utilized to train another.
Interestingly, the adoption of Chinese AI models is gaining traction in the Global South. For instance, AI Singapore, a government-backed initiative, opted for Alibaba’s Qwen over Meta’s Llama as the foundation for its latest regional model. Meanwhile, Malaysia announced plans last year to base its domestic AI ecosystem on solutions provided by DeepSeek. Entrepreneurs in cities such as Nairobi, São Paulo, and San Francisco are also increasingly building services inspired by Chinese models.
In contrast, U.S. companies continue to adhere to closed strategies, citing the necessity to recover substantial training costs and concerns about potential misuse of their models. Chinese firms, however, are actively incorporating external feedback and contributions into their development processes, especially as U.S. export controls limit access to advanced computing hardware. This evolving dynamic suggests that open-source models are not only transforming the technological landscape but also fostering a more multipolar future for AI, one that diverges significantly from the expectations held in Silicon Valley.
See also
SpaceX Targets $60B Cursor Acquisition to Boost AI Coding Power Ahead of IPO
AI Video Landscape Shifts as Grok, Runway, and Kling Emerge Post-Sora Fallout
TSMC Projected to Double Revenue by 2030, Positioning as AI Chip Leader Next to Nvidia
OpenAI Briefs Governments on GPT-5.4-Cyber’s Capabilities for National Security
Amazon’s $100 Billion Anthropic Deal Fuels 115% Growth for Astera Labs and 201% for Credo
