中国AI模型占主流开发者平台过半用量 引发美方关切

Overall, Chinese AI models have entered the mainstream of global enterprise procurement, driven by a combination of high performance and low prices. Individuals and companies choosing AI applications can adjust their selection strategy accordingly.

1. Key data: On OpenRouter, the token share of Chinese models rose from 6%-13% in February to 57%-67% in September; on Vercel, the share rose from 11% in January to 55% in August. Chinese models account for 67% of tokens consumed by enterprises in the Global South.

2. Core advantages: Their performance has crossed the threshold for business usability, coding capability has improved by leaps, and pricing is far below comparable U.S. models. Once performance requirements are met, price differences become the primary factor in enterprise selection; only extremely complex tasks still favor frontier U.S. models.

3. Practical guidance: For most application scenarios, enterprises can prioritize Chinese open-source models to save costs; if task complexity is very high and sensitivity to frontier capability is strong, consider premium U.S. models.

4. Additional reference: U.S. companies such as Airbnb, DoorDash, and Shopify have already integrated Chinese models. At the same time, watch for U.S. investigations, data-security disputes, and possible future model-licensing restrictions.

Chinese AI models' global expansion offers direct implications for brand owners in market positioning, pricing strategy, and channel selection.

1. Brand positioning and marketing: Chinese models entered the market through a combination of high cost-effectiveness and open source, then used price advantages to win users once performance was adequate. Brand owners can apply a similar logic, entering incremental markets with cost advantages.

2. Channel building: Developer platforms such as OpenRouter and Vercel are key channels through which Chinese models reach global enterprises. The fact that companies like Airbnb and DoorDash adopted Chinese models through platforms shows that third-party platforms are an effective route for rapid distribution.

3. Pricing and competition: Chinese models are priced far below comparable U.S. products, and this significant price gap drives enterprise adoption. In response, OpenAI and Anthropic have launched lower-priced models, indicating that price competition is intensifying.

4. Consumer trends and behavior: Global South companies show the highest usage of Chinese models, suggesting that price-sensitive markets prioritize value for money. Some complex tasks still favor U.S. frontier models, revealing clear market segmentation.

The rapid rise in Chinese AI models' global share brings sellers both incremental market opportunities and collaboration opportunities, as well as policy and compliance risks.

1. Growth markets: Entrepreneurs and government entities in Southeast Asia, Latin America, and Africa—the Global South—increasingly choose Chinese open-source models as their first option, expanding demand for related AI services and supporting products.

2. Policy impact: The U.S. has launched impact investigations and made allegations around distillation and data leakage. Sellers involved in taking Chinese models overseas need to watch export controls, data security, and trade-secret risks.

3. Opportunities and models: The open-source ecosystem enables cloud services, hardware support, and integrated solutions; Huawei, ZTE, and others are already seeing incremental market growth. The World AI Cooperation Organization has attracted 29 countries, and the open, customizable ecosystem creates a window for collaboration.

4. Recommendations: Use the current window of usable performance and low prices to capture demand from the Global South, while tracking regulatory moves that may introduce tiered capability licensing and preparing compliance arrangements early.

The rise of Chinese AI models gives manufacturers a dual opportunity: lower-cost AI adoption and participation in the hardware ecosystem.

1. Production and design needs: Chinese models' coding capabilities and agent support have improved across the board, enabling applications in product R&D, industrial software, and automated workflows. Factories can now access AI capabilities at lower cost.

2. Business opportunities: The expanding open-source ecosystem is driving demand for cloud services, hardware support, and supporting solutions. Huawei and ZTE are already capturing incremental market gains. Factories can position themselves as hardware or solution suppliers within the AI ecosystem.

3. Digital transformation insights: Platform call data shows strong global demand. Factories can use low-cost AI to improve digital transformation and build service capabilities around open-source models.

4. Risks and direction: The open-model path remains constrained by chip export controls and capital-market restrictions. Future governance may introduce tiered licensing systems, so factories need to monitor supply-chain stability and compliance risks.

The trend toward low-cost, open-source Chinese AI models is reshaping the AI services market. Service providers need to strike a new balance among technology, cost, and compliance.

1. Industry trends: Chinese models' share of calls on OpenRouter and Vercel has risen sharply, with Global South enterprises showing the highest usage. Their combined performance and cost advantages make Chinese models a default option for enterprises.

2. New technologies: Chinese open-source models now can stably support advanced agent applications, and coding capabilities have improved dramatically. Model distillation remains controversial, but it also represents a low-cost capability-optimization approach.

3. Customer pain points: Customers face trade-offs between performance and cost, limited compute access due to U.S. chip export controls, and concerns about model data security and leakage of trade secrets.

4. Solutions: Service providers can build deployment, cloud, tuning, and compliance assessment offerings on Chinese open-source models, helping clients capture the price advantage. They should also retain access to U.S. frontier models for extremely complex tasks while mitigating geopolitical risk.

The high penetration of Chinese AI models creates new challenges for AI development platforms around model richness, merchant operations, and compliance management.

1. Platform demand and data: OpenRouter and Vercel users have shown surging demand for Chinese models, with token shares reaching 57%-67% and 55%, respectively. Platforms need to integrate and optimize the display of Chinese models promptly.

2. Latest practices: Both OpenRouter and Vercel use platform data to show model consumption trends and help users compare performance and price. Platforms can use similar data-driven operations to guide model selection.

3. Merchant recruitment and operations: Both Global South enterprises and U.S. companies are heavily using Chinese models. Platforms can attract more Chinese models to enrich their ecosystems while establishing tiering, benchmark testing, and transparent pricing mechanisms.

4. Risk avoidance: In light of U.S. investigations, concerns about compute circumventing export controls, and accusations of distillation and data leakage, platforms need compliance review processes and strict permissions controls over model calls and data transfer.

The surge in overseas adoption of Chinese AI models reveals a new pattern of China-U.S. AI divergence and a reshaping of the global AI industry.

1. New industry dynamics: Chinese models' share on OpenRouter rose from 6%-13% in February to 57%-67% in September, and on Vercel from 11% in January to 55% in August. Sixty-seven percent of token consumption by Global South enterprises comes from Chinese models, demonstrating the market impact of the open-source approach.

2. New issues: U.S. export controls have gaps—Chinese companies can access compute through overseas data centers. Distillation dependence cannot be reliably measured. U.S. allegations that DeepSeek and others forwarded requests for training have ignited data-security disputes; China denies them.

3. Policy and regulatory implications: The analysis suggests regulators may adopt a tiered licensing system for model openness to balance national security and technology exports. The China-U.S. meeting only established a dialogue mechanism; mutual trust remains a major obstacle.

4. Business models: China's World AI Cooperation Organization (29 countries) and the U.S.-led Pax Silica alliance (more than 20 countries plus the EU) represent two parallel systems. A future structure may emerge in which the U.S. captures major profits while China captures a massive user base; companies and users will choose based on cost and performance trade-offs.