American restrictions on China’s access to advanced artificial-intelligence technology are prompting a response in Beijing that extends far beyond the substitution of imported semiconductors. China is combining technological self-reliance, industrial deployment, regulatory control, open-model distribution and diplomatic outreach into a coordinated strategy designed to reduce its exposure to American pressure while expanding its influence over the emerging global AI economy.
Washington’s increasingly hawkish position rests on the belief that leadership in advanced computing will confer decisive economic, military and geopolitical advantages. Export controls have consequently expanded beyond individual semiconductor products to encompass manufacturing equipment, technical expertise, investment and the international distribution channels through which restricted chips might reach Chinese customers. Recent efforts to strengthen scrutiny of Asian intermediaries demonstrate that enforcement is moving from rules written in Washington towards the direct policing of global supply chains.
China’s response has five principal dimensions. Beijing is accelerating the development of indigenous chips, computing infrastructure and software. It is prioritising efficiency and industrial applications in areas where China’s manufacturing scale may compensate for limitations at the technological frontier. Chinese companies are using open-weight models to build international adoption and challenge the more proprietary business models of leading American laboratories. The Chinese government is tightening domestic control over AI while presenting itself internationally as an advocate of openness and inclusive development. Finally, China is treating AI governance, technical assistance and infrastructure exports as instruments of foreign policy, particularly in its relations with developing economies.
The result is not straightforward technological decoupling. Chinese laboratories still depend, directly or indirectly, on important elements of the international semiconductor ecosystem, while American companies and researchers increasingly encounter Chinese open models in global software and research communities. Instead, the two countries are constructing competing technological spheres that remain partially intertwined.
From Selective Controls to Technological Containment
American restrictions on Chinese technology originated as targeted national-security measures, but they have evolved into a wider effort to constrain China’s ability to develop and manufacture advanced computing systems. The export-control regime introduced in October 2022 restricted high-performance chips, semiconductor-manufacturing equipment and assistance provided by American personnel to certain Chinese facilities. Subsequent measures have sought to close loopholes, prevent diversion through third countries and extend scrutiny across the wider AI supply chain. Details of the original controls and their later revisions are available from the US Bureau of Industry and Security .
The strategic argument is that advanced AI has dual-use characteristics. The same computing systems that train commercial language models can support intelligence analysis, cyber operations, autonomous weapons and military planning. Restricting Chinese access to frontier computing is therefore presented in Washington not simply as industrial protection, but as an attempt to slow the modernisation of a strategic competitor.
The chart above shows exports of integrated circuits to China (which includes some kinds of GPU chips, although Nvidia ones are not singled out). While demonstrating clearly the impact of the 2022 export controls, it also shows that the policy has become less predictable. Rules governing which chips may be sold, which customers qualify for licences and what conditions should accompany exports have changed repeatedly. This volatility increases compliance costs for multinational companies and encourages both Chinese customers and third-country suppliers to assume that access to American technology may eventually be withdrawn regardless of formal compliance.
Greater scrutiny of Nvidia’s authorised Asian distribution network illustrates the shift towards enforcement outside China itself. Buyers in Singapore, Malaysia, Japan and other markets have faced pressure to demonstrate the location and ultimate use of data centres and computing equipment. The issue became particularly important in Washington after strong growth in exports of some electronic equipment to east Asia following the export bns in China, as seen in the charts below. U.S. exports of computers and processing units to Asia have shifted sharply away from China since 2022. Once the leading destination, China has declined since 2018, while Japan, Singapore, Malaysia and Taiwan accelerated from 2024. Taiwan emerged as the largest market in 2025, surpassing 5 million in reported export value that year.
Measures to prevent these markets becoming a conduit to China may make diversion more difficult, but they also internationalise the costs of the US-China dispute and place commercial partners under pressure to participate in American technology controls. This trend has been examined in reporting by the Financial Times.
From Beijing’s perspective, these policies reinforce a long-standing conclusion: dependence on American technology constitutes a strategic vulnerability. Export controls are consequently interpreted not as temporary bargaining instruments but as evidence that the United States intends to preserve a structural hierarchy in which China remains dependent on Western intellectual property, equipment and standards.
Self-Reliance Through State-Directed Industrial Policy
China’s first major response is an accelerated campaign for technological self-sufficiency. This includes subsidies, government procurement, investment funds, research programmes and support for domestic producers of processors, memory, semiconductor equipment, cloud infrastructure and industrial software.
Self-reliance should not, however, be confused with immediate self-sufficiency. China remains constrained by limited access to the most advanced lithography systems and other specialised equipment. Producing a chip capable of running AI workloads is not equivalent to producing it economically, reliably and at sufficient scale. Yield rates, energy efficiency, memory bandwidth, interconnects and software compatibility are as important as headline processing power.
China is therefore pursuing redundancy rather than a clean break from global supply chains. Where advanced imported chips remain available, companies have strong incentives to acquire them. Where access is restricted, laboratories can redesign models, distribute training across larger numbers of less capable processors, stockpile components or use remote computing infrastructure. At the same time, Chinese chipmakers are being encouraged to develop alternatives that may initially be less efficient but are adequate for domestic deployment.
This layered response reduces the likelihood that any single restriction will halt Chinese AI development. Research on hardware-centred controls suggests that laboratories can partly compensate for weaker chips through improved algorithms, model architectures and engineering practices. Restrictions still impose meaningful costs, particularly at the frontier, but the relationship between computing inputs and model performance is not fixed. One analysis of these limitations is presented in the research paper “Whack-a-Chip: The Futility of Hardware-Centric Export Controls” .
American pressure may consequently alter the direction of Chinese innovation. Instead of attempting to reproduce the capital-intensive strategies of leading US laboratories exactly, Chinese developers have stronger incentives to reduce inference costs, compress models, improve the use of available memory and optimise systems for locally produced hardware.
China’s appetite for foreign technology grew sharply over the decade, led by integrated circuits, whose imports climbed from roughly 231 to 386 in the chart’s units after peaking in 2021. Purchases of chip-making machinery nearly quadrupled to a 2024 record, while imports of optical inspection equipment surged from a low base, underscoring Beijing’s push to strengthen domestic semiconductor production capacity. Yet such imports should also be seen as building up China's own production base, supplementing internal production to ensure that it can create the compute required to train its own AI models.
This is strategically important. A Chinese model that performs slightly below an American frontier system but costs substantially less to deploy may be more commercially significant across manufacturing, logistics, public administration and consumer services. Technological leadership is not determined solely by which company produces the most capable model. It also depends on which country can incorporate AI into the largest number of economically useful processes.
“AI Plus” and the Advantage of Industrial Scale
China’s second response is to turn AI competition from a race between foundation models into a contest over industrial transformation. The government’s “AI Plus” agenda calls for artificial intelligence to be integrated across manufacturing, transport, energy, healthcare, education, agriculture and public services.
This reflects one of China’s strongest comparative advantages. The country possesses dense manufacturing clusters, extensive digital-payment systems, large logistics networks, substantial engineering capacity and a domestic market capable of supporting rapid experimentation. Factories can combine AI with industrial robots, machine vision, predictive maintenance and automated quality control. Vehicle manufacturers can integrate models into autonomous-driving systems, while logistics companies can apply them to routing, warehousing and demand forecasting.
The policy also creates demand for domestic computing products. Chinese processors may struggle to compete directly with the best American chips in frontier training, but they can become commercially viable when purchased by state-owned enterprises, local governments, universities and manufacturers. Deployment produces revenue, operational data and engineering experience, supporting subsequent generations of domestic technology.
China’s approach therefore links AI policy to its wider advanced-manufacturing strategy. The objective is not only to create successful software companies, but to raise productivity, strengthen supply chains and establish Chinese firms as providers of integrated AI-enabled industrial systems.
Major Chinese technology exhibitions have offered a concentrated display of this strategy. Companies increasingly present foundation models alongside robotics, industrial hardware, autonomous systems and consumer applications. Although demonstrations do not necessarily prove commercial readiness, they highlight Beijing’s intention to compete through the combination of software, manufacturing and large-scale implementation. The broader direction of this competition has been covered by The Wall Street Journal .
This approach gives China a different risk profile from the United States. American AI investment is heavily concentrated in a small number of frontier laboratories, hyperscale cloud companies and semiconductor designers. China’s model is more widely distributed across national ministries, local governments, state enterprises, universities and private technology firms. Such diffusion can generate waste and duplication, but it also makes the national effort difficult to disrupt through sanctions directed at individual companies.
Open Models as a Commercial and Geopolitical Instrument
Perhaps China’s most consequential response to American pressure is its embrace of open-weight AI. Chinese developers have increasingly released models that can be downloaded, adapted and deployed by outside organisations. This contrasts with the closed systems operated by several leading American laboratories.
Open-weight distribution serves multiple purposes. It encourages external developers to build applications around Chinese models, expands their presence in international research and lowers adoption costs in markets that cannot afford premium American services. It also allows companies and governments to run models on their own infrastructure, an important consideration in countries concerned about data sovereignty or dependence on US cloud platforms.
Research published in 2026 argues that American technology restrictions may have unintentionally accelerated China’s open AI ecosystem. As access to frontier hardware became less certain, open and locally adaptable software acquired greater strategic value. Chinese participation in open-model repositories increased, while Chinese-origin systems spread through research and commercial experimentation outside China. This argument is developed in “US Policies Unintentionally Accelerated China’s Open AI Ecosystems” .
The emergence of increasingly capable Chinese open models underscores this development. Releases from companies such as Alibaba, DeepSeek and Moonshot AI have been presented as competitive with advanced American systems on selected tests, although company benchmarks should be treated cautiously and independently verified. Their significance lies less in isolated rankings than in the combination of improving performance, relatively low prices and permissive availability. Recent Chinese open-model developments have been examined by The Verge .
Open models can function as a form of infrastructure diplomacy. A government or company that adapts a Chinese model, trains personnel to use it and integrates it with Chinese cloud or telecommunications systems becomes part of a wider technical ecosystem. Even when the underlying software is freely available, associated services such as hosting, customisation, hardware, maintenance and cybersecurity can create enduring commercial relationships.
There are limitations. Open weights do not necessarily mean transparent training data, unrestricted use or freedom from political controls. Chinese models are generally developed within a regulatory environment requiring adherence to state-defined content rules. Countries adopting them must assess security, censorship, privacy and dependency risks just as they would with proprietary American systems.
Nevertheless, the open-model strategy complicates Washington’s containment efforts. Hardware can be intercepted at borders and cloud services can be regulated, but software distributed internationally through developer communities is considerably more difficult to contain.
Regulation at Home, Openness Abroad
China’s international messaging emphasises cooperation, accessibility and opposition to technological monopolies. Domestically, however, Beijing maintains one of the world’s most interventionist AI-governance systems.
Chinese regulations cover recommendation algorithms, synthetic media, generative-AI services, data security, model registration and content labelling. Providers serving the public must address politically sensitive content, protect personal information and comply with state-security requirements. Regulatory agencies have preferred a series of targeted rules over a single comprehensive AI law, allowing the state to adjust requirements as new applications emerge. Relevant measures and policy guidance are published by the Cyberspace Administration of China .
This arrangement reflects a fundamental Chinese policy objective: AI should be innovative enough to support economic growth but sufficiently controllable to avoid political instability. Beijing is prepared to promote private competition, but it does not accept the proposition that developers should determine the social and political boundaries of AI systems independently of the state.
Hawkish American policy strengthens the national-security justification for this control. Restrictions on chips, investment and software enable Chinese officials to portray technology governance as part of a wider struggle for sovereignty. Data localisation, security reviews and scrutiny of cross-border acquisitions can be defended as necessary protections against foreign coercion.
The tension between domestic control and international openness is not necessarily seen in Beijing as contradictory. China’s diplomatic position is that states should be free to govern AI according to their own political systems and development priorities. “Openness” therefore means access to technology and opposition to American exclusion, rather than the absence of government regulation.
Building an Alternative International AI Order
China’s final response is diplomatic. Beijing increasingly presents the dispute as a choice between American technological exclusivity and Chinese-backed inclusive development. This message is directed particularly towards countries in Asia, Africa, Latin America and the Middle East.
Chinese officials have criticised restrictions on technology sharing and called for wider international cooperation. Beijing has proposed training programmes, technical assistance and governance mechanisms aimed at developing economies. It has also promoted international institutions intended to provide an alternative or complement to arrangements dominated by the United States and its allies. China’s evolving international AI strategy has been analysed by the Carnegie Endowment for International Peace .
The proposition is attractive to governments that want AI capabilities but lack domestic computing infrastructure. China can offer combinations of telecommunications equipment, cloud services, data centres, smart-city systems, training and financing. Its ability to package these elements may matter more than whether each individual component is technologically superior to its American equivalent.
The diplomatic strategy also enables Beijing to recast export controls as a development issue. Washington presents restrictions as narrowly targeted national-security measures. China portrays them as attempts to monopolise innovation and deny developing countries access to advanced technology. This framing may resonate where governments have previously experienced restrictions on finance, telecommunications or dual-use technology.
American policymakers are increasingly aware of this vulnerability. Critics of a predominantly defensive US strategy argue that export controls must be accompanied by financing, infrastructure partnerships and more accessible American technology for trusted countries. Without a positive offer, restrictions may slow Chinese development at the frontier while leaving Beijing free to gain influence in the wider global market. The debate over a more internationally competitive American strategy has been discussed by the Washington Post .
Commercial and Strategic Outlook
The most likely outcome is neither complete Chinese technological independence nor the indefinite preservation of uncontested American leadership. The competition will instead become segmented.
The United States is likely to retain important advantages in advanced semiconductor design, manufacturing equipment, frontier computing infrastructure and leading proprietary models. China will remain under pressure where performance depends on access to the most advanced hardware and globally specialised supply chains.
China, however, may establish advantages in cost-efficient models, open-weight distribution, industrial robotics, AI-enabled manufacturing and deployments tailored to emerging markets. The combination of large-scale domestic adoption and international infrastructure partnerships could allow Chinese technical standards to spread even without clear superiority at the frontier.
For multinational businesses, four risks require particular attention. The first is regulatory fragmentation: systems authorised in one country may be restricted in another. The second is supply-chain exposure, including the possibility that chips, cloud services or software updates become unavailable following a political decision. The third is data governance, as companies may be required to separate Chinese and international datasets and computing environments. The fourth is reputational and national-security scrutiny associated with adopting models originating in either strategic bloc.
Companies should therefore plan for technological optionality. This may include maintaining multiple model providers, designing applications that can migrate between clouds, documenting the provenance of training data and components, and undertaking enhanced due diligence on distributors and end users. Boards should treat AI procurement as a geopolitical decision rather than a conventional software purchase.
Conclusion
China’s response to a more hawkish American AI policy is best understood as strategic adaptation under constraint. US controls have raised the cost of Chinese development and exposed genuine weaknesses in the country’s semiconductor ecosystem. They have not, however, halted progress.
Instead, pressure is encouraging Beijing to mobilise industrial policy, expand domestic procurement, improve computational efficiency and accelerate the use of open models. China is pairing these measures with a diplomatic campaign that presents its technology as accessible to countries excluded from the most advanced Western systems.
The central paradox for Washington is that controls can simultaneously weaken and strengthen China. They can delay access to frontier equipment while strengthening political support for self-reliance. They can constrain individual laboratories while encouraging the diffusion of cheaper open systems. They can protect American technology while making Chinese alternatives more attractive to governments seeking greater autonomy from the United States.
For China, the principal difficulty will be balancing control with innovation. Extensive political and regulatory supervision may reduce social risks from Beijing’s perspective, but it can also discourage experimentation, restrict access to information and limit international trust. China’s strategy will succeed only if domestic chip development advances, industrial applications generate productivity gains and foreign partners perceive Chinese systems as useful rather than politically burdensome.
The emerging AI order will therefore not be defined solely by which country trains the most powerful model. It will be shaped by which country can build the most resilient supply chain, achieve the broadest deployment, offer the most attractive international partnerships and persuade other states to adopt its standards. On those measures, the United States retains formidable advantages, but China’s response demonstrates that technological pressure is producing a more determined and increasingly sophisticated competitor.
Sources and Further Reading
1. The White House, “Promoting Advanced Artificial Intelligence Innovation and Security”, June 2026.
2. State Council of the People’s Republic of China, Policy guidance on the implementation of China’s “AI Plus” initiative, August 2025.
3. Government of the People’s Republic of China, “China Unveils Guidelines to Regulate and Boost Innovative AI Agents”, May 2026.
4. Cyberspace Administration of China, Policy and regulatory guidance concerning artificial intelligence, May 2026.
5. Carnegie Endowment for International Peace, “China’s Pivot on Global AI”, May 2026.
6. Council on Foreign Relations, “How Trump Should Approach AI Talks With China: Targeted Dialogue, Maximum Pressure”.
7. War on the Rocks, “China’s AI Governance Offensive Threatens US Technology Leadership”.
8. American Affairs Journal, “National Orchestration and Provincial Competition: China’s Industrial Policy for AI Dominance”, May 2026.
9. BBC News, “China Is Winning One AI Race, the US Another — but Either Might Pull Ahead”.
10. Deutsche Welle, “China’s New Rules Give the West a New Headache”.
11. ScienceDirect, “Small, Swift and Effective Legislation in China: Towards Adaptive Artificial-Intelligence Regulation”, 2026.
12. China Economic Net, “China Moves Forward With Its AI Plus Initiative”, November 2025.
13. China Daily, “China Issues Guidelines to Regulate and Promote AI Agents”, May 2026.
14. Xinhua News Agency, Report on the development and application of artificial intelligence in China, September 2025.
15. South China Morning Post, Artificial-intelligence policy coverage and analysis.
16. Fudan Development Institute, Research and analysis concerning international artificial-intelligence policy, 2026.
17. China-US Focus, “How China and the US Can Expand Artificial Intelligence Cooperation”.
18. Centre for Technology Research, “US-China AI Governance Dialogue”.
19. Geopolitical Monitor, “AI With Chinese Characteristics”.
20. Baidu AI Legal Research, Analysis of China’s national artificial-intelligence development and regulatory framework.
