Hardware vs. Software: How US Chip Controls and China's Proposed Model Export Rules Are Splitting the Global AI Ecosystem

The United States and China are pursuing fundamentally different strategies for controlling the spread of advanced artificial intelligence. Since October 2022, Washington has worked to deny Beijing access to the advanced semiconductors needed to train and run frontier AI models, according to a report by Gregory C. Allen and Isaac Goldston of the Center for Strategic and International Studies. China, by contrast, is now drafting export controls aimed at the models themselves — including open-weight releases that anyone can download and run. For organizations that build on or deploy open-weight AI, the practical consequence is a compliance landscape that may soon require tracking two parallel regimes: one governing chips, one governing model weights.

The Diverging Landscape of AI Export Controls

The two approaches reflect different anxieties. US policy, as described by the Peterson Institute for International Economics' Martin Chorzempa, is designed to prevent adversaries from obtaining frontier AI capabilities that could boost commercial competitiveness and military strength, and to stop competing AI companies from extracting hard-won capabilities from leading US models. China's draft rules, reported in October 2026, would instead give Beijing a lever over the outward diffusion of its own AI output — the open-weight models that Chinese labs have released freely to date. Dean W. Ball, head of strategic futures at OpenAI, characterized Beijing's framework as a "mirror image" of the model-weight export controls the US government has already put in place.

US Strategy: Restricting Access to Advanced AI Hardware

The American approach targets physical technology. As the CSIS report by Allen and Goldston details, the United States has devoted significant resources since October 2022 to restricting China's access to AI and advanced semiconductor technologies, with the Department of Commerce issuing four additional far-reaching export control updates in the final months of the Biden administration. Chorzempa notes that the pre-2025 US approach focused on depriving China of the chips needed to train and run advanced AI models, including by denying those chips to countries deemed at risk of diverting technology to China. The strategy also depends heavily on allied cooperation, since advanced chips and the tools to make them pass through other jurisdictions.

China's Proposed Controls: Targeting AI Models and Software

China's draft rules, reported by The Decoder's Matthias Bastian, would tighten export controls on open-weight models and 23 key AI technologies. The framework reportedly uses public rankings from domestic and foreign evaluators and industry platforms — measuring overall performance, specific tasks, safety, or global usage — to determine which models fall under control, with top-10 status appearing to be a trigger. Ball's assessment is pointed: "This move does not mean 'China is shutting down open-weight models,' but it does indicate the Chinese government establishing the optionality to shut down future open-weight models at will." A spokesperson for China's Ministry of Commerce told the South China Morning Post that licensing requirements for open-source AI technologies would be expedited, with immediate clearance upon filing — though analysts say the practical meaning of that process remains unclear.

US vs. China: Diverging AI Export Control Strategies

The table below contrasts the two regimes on their regulatory focus, targets, stated goals, and likely consequences for open-weight models. The comparison draws on the CSIS account of US policy, the Peterson Institute's analysis of US aims, and reporting on China's draft framework.

US and Chinese AI export control strategies compared
Dimension	United States	China (Proposed)
Regulatory focus	Controlling access to advanced physical hardware — specifically advanced semiconductors and AI chips.	Controlling the export of AI model software and related technologies, including open-weight models and 23 key AI technologies.
Primary target	Advanced semiconductors and AI chips; controls began in October 2022, with four further Commerce Department updates in late 2024.	Open-weight AI models and other key AI technologies; overseas transfers may require registration or licensing, with top-ranked models apparently in scope.
Stated goal	Prevent adversaries from accessing frontier AI capabilities that could boost commercial, military and strategic advantage, and prevent competitors from distilling US AI capabilities.	Establish government optionality to control the diffusion of China's advanced AI capabilities — the ability to shut off future open-weight releases at will, rather than an outright ban.
Potential impact on open-weight models	Indirect: restricting the chips used to train and run advanced models, including by denying chips to countries deemed diversion risks.	Direct: additional hurdles for obtaining or deploying Chinese open-weight models overseas, affecting companies hosting models for customers outside China and future model releases.

The structural contrast is the point: one regime governs inputs to AI development, the other governs outputs. A company can comply with US chip rules and still find its model supply chain disrupted by Chinese weight-export rules, and vice versa.

Impact on the Global Open-Weight AI Ecosystem

Chinese open-weight models have become globally significant. Stanford HAI's analysis of China's open-weight ecosystem warns that widespread global adoption of Chinese models may reshape global technology access and reliance patterns, and affect AI governance, safety and competition — while cautioning that government support for open-weight development, though substantial, carries no guarantee of continuing. The proposed export controls would formalize that uncertainty. Business Standard's reporting, citing an SCMP account of analyst Alex Colville of the Australian Strategic Policy Institute, notes that new compliance requirements could delay or prevent Chinese AI companies from partnering with global cloud providers to deploy their most capable open-weight models overseas. Colville added that the practical meaning of the proposed expedited licensing process remained unclear, and that the extent of the impact will depend on the final rules.

Two boundaries matter for organizations already using Chinese models. First, the proposed rules do not automatically require developers to stop using Chinese models they have already downloaded, according to Business Standard's reporting. Second, the risk concentrates on future releases: if overseas access to some advanced Chinese models becomes subject to licensing, developers could face additional hurdles obtaining or deploying them, narrowing the range of models available to businesses — including those that want to run AI systems locally.

Strategic Shifts and Compliance Risks for Organizations

Chorzempa raises a scenario that cuts against the open-weight ecosystem from another direction: nonstate actors might use powerful Chinese open-weight models, fine-tune the guardrails out of them, and engage in cyberattacks — including against China — and the inability to monitor and shut off access to open models applies to criminals as well. He also notes that Chinese labs may shift away from open-weight models for economic reasons, since building a sustainable business around giving away the fruits of expensive training runs is difficult.

Lawfare's reporting adds context on why openness has persisted: Z.ai's Director of Product Zixuan Li told the ChinaTalk substack in November 2025 that companies release open-weight models to contribute to research and build acceptance overseas, saying, "I think it is necessary to be open right now for people to use our models." Beijing, for its part, has promoted AI as a global public good — a position the new draft controls would qualify with export licensing.

For organizations, the exposure is concrete: cloud partnerships involving Chinese models, locally deployed open-weight systems, and downstream products fine-tuned from Chinese base models could each fall within a licensing perimeter that does not yet exist in final form. The analysis here rests on publicly available reporting and policy analysis rather than the draft regulations' official text, and it is not legal advice on export control compliance.

Navigating the New AI Regulatory Landscape

Organizations developing or utilizing open-weight AI models must assess their exposure to both US hardware controls and China's proposed model controls, adapting supply chain and collaboration strategies accordingly. That means inventorying which open-weight models underpin current products, identifying whether any are Chinese in origin and how they were obtained, and reviewing cloud or hosting partnerships that move Chinese models across borders. The measurable indicator to watch is the finalization and enforcement details of China's AI export regulations, together with any new US policy updates — particularly the scope of controlled technologies and the licensing requirements. Until those final rules land, the expedited-clearance promise from China's Ministry of Commerce remains an untested process, and the practical weight of the "mirror image" will stay unknown.