中美AI全球博弈升级 开放权重模型成竞争焦点

This article maps the core timeline of how the U.S.-China AI rivalry escalated into competition over the open-weight model segment around 2026, alongside actionable, practical insights for understanding the AI landscape.

1. Key developments overview: At the 2026 World Artificial Intelligence Conference in Shanghai, Chinese firm Moonshot AI released Kimi K3, an open-weight model on par with leading U.S. offerings. China also launched the World Artificial Intelligence Cooperation Organization to promote global sharing of AI technology. The U.S. had previously established Pax Silica, an exclusive AI and semiconductor alliance, and later pressured allied nations not to join China’s cooperation framework. While tensions escalated between the two sides in September, room for commercial cooperation remained intact.

2. Practical takeaways: First, Chinese open-weight models are 60%–90% cheaper to use than U.S. closed-source models. They support downloading, modification, and local deployment, and enable data to be retained on domestic servers. Second, Chinese open-weight models now account for 58% of enterprise API calls on U.S. developer platforms, and were previously deployed by open-source platform Hugging Face for cybersecurity forensics. Third, most countries around the world select AI tools primarily based on cost, security, and compatibility, rather than aligning with ideological narratives.

The global AI race has settled into a rivalry between two differentiated technology paths led by the U.S. and China, and the growing adoption of open-weight models opens a new viable path for brands to deploy intelligent applications and expand into global markets.

1. Technology selection guidance: Leading Chinese open-weight models cost 60%–90% less to use than U.S. closed-source models, support local deployment and domestic data residency, and can help brands substantially cut the cost of intelligent transformation while mitigating compliance risks from overseas service outages and cross-border data transfers. The U.S. previously suspended overseas access to Anthropic’s models without warning, disrupting critical business operations for users in multiple countries; such risks can be avoided by adopting open-weight models.

2. Market trend reference: The share of enterprise API calls to Chinese open-weight models on U.S. developer platforms rose from 1.2% at the end of 2024 to 58% in July 2026. When selecting AI vendors, emerging markets in Southeast Asia, Africa, and other regions prioritize cost and autonomous control. For brands expanding into these markets, intelligent hardware products adapted to open-weight AI tools and integrated with China’s supply chain advantages are better positioned to meet local demand.

3. Product development note: Open-weight models release all training parameters and support secondary modification, allowing brands to build customized intelligent applications on top of these models and reduce reliance on overseas closed-source vendors.

The open-weight model segment is rising rapidly amid U.S.-China AI competition, creating market opportunities for cross-border and AI-focused sellers from policy misalignment, while requiring proactive mitigation of geopolitical compliance risks.

1. Opportunity highlights: On one hand, Chinese open-weight models cost 60%–90% less than U.S. closed-source models. Emerging markets in Southeast Asia, Africa, and other regions prioritize cost and autonomous control when procuring AI solutions, creating clear growth potential for cost-effective intelligent hardware compatible with AI models, as well as vertical-scenario AI applications. On the other hand, leading U.S. cloud providers including Microsoft, Amazon, and Google are in talks to host China’s Kimi K3 model. If launched, this will create new revenue-sharing collaboration channels, and relevant sellers can position themselves for the associated ecosystem early.

2. Risk alerts: The U.S. is currently considering a comprehensive ban on domestic use of Chinese open-weight models, and has accused six Chinese AI companies of engaging in so-called "malicious distillation." Sellers operating in the U.S. market must ensure compliance to avoid risks such as account penalties.

3. Reference operating model: Open-weight models support local modification and deployment, allowing sellers to build lightweight vertical applications on top of them, reducing R&D costs while aligning with data security requirements across different countries.

The global spread of the open-weight model technology path amid U.S.-China AI competition provides clear direction for manufacturing factories to advance digital transformation and capture the hardware export dividend driven by AI adoption.

1. Product design and business opportunities: China holds a dominant position in the physical manufacturing supply chain for products including robots. As open-weight AI models gain global traction, intelligent hardware and robot products designed for local deployment will see an export growth window. Emerging markets in Southeast Asia, Africa, and other regions have strong demand for cost-effective intelligent hardware that supports local data residency as they build out their AI capabilities. Factories can design products tailored to open-weight models, and also pursue B2B orders by partnering with developing country projects under the World Artificial Intelligence Cooperation Organization framework.

2. Digital transformation takeaways: Open-weight models cost 60%–90% less than U.S. closed-source models and support local deployment. When implementing intelligent production and digitalized e-commerce operations, factories can use these models to build proprietary intelligent systems, lowering transformation costs while avoiding risks of production data leakage and production disruptions from overseas service outages. The U.S. previously cut off overseas access to its models without warning, halting critical operations in multiple countries; such risks can be effectively mitigated through local deployment.

The global AI industry has formed a competitive dynamic between two technology paths: U.S.-led closed-source models and China-led open-weight models. Technology service providers can capture industry growth dividends by building solutions that address customer pain points in the open-weight segment.

1. Industry trend assessment: Leading Chinese AI models have adopted the open-weight path, offering 60%–90% lower usage costs than U.S. closed-source models and supporting local modification and independent deployment. As of July 2026, Chinese models account for 58% of enterprise API calls on U.S. developer platforms. Open-weight models are set to become one of the mainstream technology choices for global AI applications, particularly in emerging markets.

2. Core customer pain points: First, enterprises and government agencies across countries have strong data sovereignty requirements, and are concerned about risks of data outbound flows from closed-source models and service disruptions triggered by U.S. policy changes. In June 2026, the U.S. required Anthropic to suspend overseas model access, disrupting critical business operations in multiple countries. Second, small and mid-sized customers cannot afford the high costs of closed-source models. Third, customers have demand for customized development, which cannot be met by closed-source models that do not disclose parameters. For example, when open-source platform Hugging Face faced a cyberattack, it was unable to use a closed-source model to analyze malicious code due to restrictions in the model’s terms of service.

3. Solution direction: Service providers can offer local deployment, customized modification, and model operation and maintenance services based on Chinese open-weight models, with a focus on meeting AI development demand in emerging markets.

Against the backdrop of evolving U.S.-China AI competition, platform enterprises can capture ecosystem-building dividends from open-weight models while dynamically addressing compliance risks brought by geopolitical policies.

1. Ecosystem development and merchant acquisition priorities: Open-weight models disclose all parameters, support user-led secondary development and local operation, and are more conducive to fostering a diverse vertical application ecosystem than closed-source models. Platforms can prioritize recruiting developers and service providers building applications on Chinese open-weight models, with a focus on cost-effective AI services and data security and compliance applications, to serve the broad base of emerging market users and enterprise customers with data residency requirements. Leading U.S. cloud platforms are currently in talks to integrate China’s Kimi K3 model, which validates the ecosystem appeal of this category of models.

2. Operations management guidance: Global users prioritize cost, data security, and service stability when selecting AI services. U.S. closed-source models have previously experienced unexpected service outages, as well as terms-of-service restrictions that prevented users from responding to cybersecurity incidents. Platforms should implement multi-model backup in operations to avoid disruptions to normal platform functioning caused by service interruptions of a single closed-source model.

3. Risk mitigation guidance: The U.S. is pressuring allied nations not to join China’s AI cooperation framework, and is considering a ban on the use of Chinese open-weight models in the U.S. When conducting cross-border business, platforms should dynamically track regulatory policies across different regions to avoid compliance risks.

As the global U.S.-China AI rivalry escalates, open-weight models have become a core focal point of competition between the two sides, and the resulting new industry characteristics and governance issues hold significant research value.

1. New industry competition dynamics: The U.S. and China have formed a competitive landscape of differentiated technology paths. The U.S. leads the closed-source, proprietary model route, launching the Pax Silica initiative in December 2025 to build an exclusive supply chain alliance with 23 partner countries and the European Union, in an attempt to form blocs along ideological lines. China is promoting the open-weight model route, establishing the World Artificial Intelligence Cooperation Organization and aligning with the Belt and Road Initiative and BRICS cooperation frameworks, expanding its global market reach by leveraging advantages including 60%–90% lower costs than closed-source models and local deployment support that aligns with countries’ digital sovereignty demands. Chinese models now account for 58% of enterprise API calls on U.S. developer platforms.

2. Emerging industry issues: The U.S. has escalated technology competition into technology blockade, pressuring allies to exclude China’s AI cooperation framework, making unsubstantiated accusations that Chinese AI companies engage in "malicious distillation," and proposing usage bans, artificially fragmenting the global AI industry chain. However, geopolitical bloc formation has not fully translated into commercial and national cooperation dynamics. A separation of political and economic dynamics has emerged: U.S. cloud providers are in talks to integrate Chinese models, Kazakhstan participates in both cooperation mechanisms, and African countries select cooperation partners entirely based on practical needs.

3. Governance implications: Global AI governance must respect countries’ digital sovereignty and oppose the formation of exclusive blocs. The room for cooperation in future high-level U.S.-China AI dialogues, as well as the commercial sustainability of the open-weight model, warrant ongoing tracking.