China’s proposed AI export controls could reshape how Chinese technology reaches overseas customers, without cutting off ordinary access to AI services. The commercial question is not simply whether a model can be sold abroad. It is whether a transaction transfers the underlying technology, gives a foreign buyer control over it, or provides access to a service.
For software buyers, chip designers and investors, those distinctions matter more than the headline. A hosted chatbot subscription and an acquisition of the technology behind it are very different transactions. Treating them as interchangeable would produce poor procurement decisions and worse valuations.
What China Has Proposed—and What Already Applies
On October 16, 2026, China’s Ministry of Commerce and Ministry of Science and Technology published Announcement No. 48, proposing additions to the Catalogue of Technologies Prohibited and Restricted from Export. The consultation covers 40 technologies, including 23 AI technologies involving large models, agents, computing clusters and information processing. Public comments run from October 16 to November 15, 2026. As of October 19, the catalogue changes remain proposals.
However, temporary registration requirements took effect with the announcement. Exporters must register covered transfers before proceeding, including transactions already started but not completed. Relevant arrangements include transferring technology ownership abroad, licensing domestic technology for overseas use, and changes that place ownership or actual control of domestic technology with an overseas entity. Registration continues until the revised catalogue takes effect. The distinction is therefore practical: the final restrictions are under consultation, but the interim reporting obligation is not. These dates, transaction categories and requirements appear in the joint announcement on proposed technology export controls and temporary registration.
Technology Transfer Is Different From Service Access
The ministries’ stated approach draws a boundary between transferring controlled technology and supplying services built on it. Under the proposed framework, overseas local deployment of restricted AI technology would fall within technology export management. Deployment on a company’s own overseas cloud could qualify for automatic licensing or other simplified procedures.
Providing API access, app or webpage interaction, SaaS applications and lightweight deployment on local devices would fall outside the stated large-model technology export scope. The proposed approach also excludes relevant technology already in the public domain through open sourcing or academic publication. Simplified procedures are envisaged for activities including AI open sourcing, overseas chip tape-outs and manufacturing, joint research using overseas resources, and participation in international standards. These distinctions concern this technology export framework, not blanket exemptions from every applicable rule; the Ministry of Commerce’s policy clarification on AI transfers and service access sets out the intended boundaries.
For buyers, the sensible starting point is the actual delivery arrangement. “AI partnership” is a marketing description, not a useful classification of what crosses the border.
How the Proposal Could Change Global Technology Trade
Hosted services could become more attractive
If transferring underlying technology requires more review than supplying hosted access, providers may prefer subscriptions over technology licensing. Overseas customers could still obtain useful AI capabilities, but with less control over deployment, modification and maintenance.
Consider a hypothetical manufacturer choosing between a hosted inspection tool and a model installed on its own servers. The hosted option might reduce the amount of technology transferred. The local installation might offer greater operational independence. Procurement teams would need to weigh that independence against classification work, contractual conditions and possible approval delays.
Neither approach wins automatically. A cheaper subscription can become expensive if switching providers requires rebuilding an entire workflow.
Acquisitions could require closer technical review
The announcement’s treatment of ownership and control makes the technology inside a target company a priority for transaction review. A buyer should not assume that purchasing shares avoids questions that would arise in a direct intellectual property sale.
Potential consequences include longer completion schedules, conditions attached to access rights, or changes to the assets included in a deal. These are possible commercial responses, not evidence that every overseas acquisition will be blocked. Valuation work should distinguish revenue from service provision from revenue that depends on transferring technology.
Research partnerships may need clearer boundaries
Joint development agreements should identify who owns existing technology, who owns new work, and which participants receive access to each. Shared repositories and loosely defined licensing rights can make those questions harder to answer after a project has begun.
The broader issue also appears in proposed Chinese solar technology export restrictions and overseas factory investment: selling an output is not necessarily the same as transferring the knowledge needed to reproduce it. AI businesses should assess their own arrangements rather than borrow assumptions from another sector.
What Buyers and Technology Partners Should Check
A useful review starts with a technical inventory, not a country label. Buyers should establish what they receive: service access, model files, software, design documentation, deployment assistance or rights to reproduce and modify the technology.
- Define the deliverables: separate ordinary service access from transfers of technology and intellectual property rights.
- Allocate responsibility: specify who handles classification, registration and any later licensing requirements.
- Protect continuity: address delays, unavailable upgrades, migration support and termination rights.
- Review ownership: examine control changes and access rights alongside the technical assets themselves.
Contracts should avoid promises that depend on an unverified regulatory outcome. A supplier’s assurance that delivery is “compliant” is less useful than a documented explanation of the transaction, its classification and the steps required before delivery.
What to Watch After the Consultation
The most useful developments will be the final technical scope, the procedures for simplified licensing, and practical guidance on borderline deployment arrangements. Businesses should watch whether the final framework makes transactions easier to classify, rather than focus only on how many technologies appear in the catalogue.
Hardware and software exposure should also be assessed separately. The questions raised by Alibaba’s AI chip and China’s semiconductor self-sufficiency push concern a related but different part of the supply chain. Access to computing equipment does not, by itself, settle access to model technology.
The practical response is neither to assume Chinese AI exports will stop nor to dismiss the proposal as paperwork. Map the technology, identify the rights crossing the border, and build contracts around the delivery model. That provides a firmer basis for decisions than treating every AI transaction as the same trade.