China Wants Its AI to Go Global, on Its Own Terms

China’s latest draft to tighten export controls on artificial‑intelligence technology reveals a calculated effort to expand the global reach of its AI while preserving a decisive grip over how, where and by whom the technology is used. Announced on 16 October, the proposal adds 23 AI‑related items to the Ministry of Commerce’s (MOFCOM) Catalogue of Technologies Prohibited or Restricted from Export, and it opens a public comment period that runs until 15 November. Although the final list is not yet settled, the immediate requirement for pre‑export registration already signals Beijing’s intent to monitor transactions before they occur, even for deals that are already in progress.

Scope of the proposed restrictions

The draft focuses on the core components of advanced AI models—weights, source code, AI agents, chip design for AI accelerators, and the interconnect and scheduling technologies needed for clusters of more than 10,000 accelerators. Notably, the regulations exclude the chips themselves and any products that merely incorporate AI technology. MOFCOM also clarifies that offering restricted large models through APIs, software‑as‑a‑service platforms, lightweight on‑device deployments or open‑source releases falls outside the export‑control scope, although the first open‑source release of a restricted item would still be treated as an export eligible for automatic licensing.

Mechanics of the registration requirement

Under the new rules, any transfer of ownership abroad, a license for overseas use, or a change of actual controller to a foreign entity triggers the registration obligation. This applies to transactions already underway, giving regulators advance notice before control of the technology changes hands. The registration system will coexist with existing licensing requirements, allowing MOFCOM to see proposed deals before they are finalized while still enforcing the traditional licensing regime for actual exports.

Learning from the Manus episode

The draft appears to be a direct response to the controversy surrounding Manus, an AI‑agent startup that moved its headquarters to Singapore before being acquired by Meta. After the acquisition closed, China’s foreign‑investment security review office ordered the deal to be unwound and barred two Manus co‑founders from leaving the country. The new registration rule would have forced a Manus‑style sale to be reported before completion, giving regulators the chance to intervene earlier. It also signals that merely relocating a company abroad does not automatically shield its core AI assets from Beijing’s oversight.

Implications for remote computing and cloud services

The draft leaves open questions about remote access to Chinese compute resources. While renting GPUs in Shanghai keeps the hardware on Chinese soil, the registration requirement would still apply to any authorization of overseas use of listed technology. The current carve‑out for APIs and software does not explicitly address bare GPU rentals, suggesting that future rules could extend control to such arrangements, especially as China expands its own cloud capacity abroad.

Strategic motivations behind the controls

China’s rationale, as outlined by MOFCOM, is to foster AI capacity building in the Global South while retaining the ability to shape international AI governance through the World AI Cooperation Organization (WAICO). WAICO, launched in July with 29 founding governments, is not a prerequisite for technology access, but membership may favor applicants when Chinese firms seek export licences. By offering automatic licensing for chip designs and joint research, Beijing hopes to bundle its AI models with the infrastructure it builds overseas, ensuring that Chinese firms retain ownership of both the compute and the models.

Comparison with the U.S.‑led Pax Silica initiative

The Chinese approach mirrors aspects of Pax Silica, a supply‑chain compact introduced by the United States that now counts about two dozen signatories. While Pax Silica emphasizes secure hardware supply chains for minerals, semiconductors and AI infrastructure, WAICO focuses on capacity building, technology sharing and governance. Both frameworks reward partners that align with their respective strategic goals, and Indonesia—a WAICO founder—has indicated that the two initiatives are not necessarily at odds, seeing WAICO as a tool‑centric effort and Pax Silica as a broader digital‑development programme.

Potential leverage and path dependence

By exporting AI models and related services, China creates a channel for long‑term influence. Users who adopt Chinese models may become dependent on subsequent releases, training know‑how, integration tools and ongoing technical support. This dependence could translate into leverage over foreign firms and governments, especially in regions that lack the capacity to develop competitive foundation models. The narrow list of prohibited items—targeting malicious code, models trained on Chinese data, defense agents and critical‑infrastructure applications—allows Beijing to frame its controls as safeguards for human oversight while simultaneously positioning itself as a leader in shaping global AI policy through WAICO.

This article was produced with AI-assisted research and editorial support. Reporting is based on the source material cited below. Sources: The Diplomat; thediplomat.com; Global1.News (29 October 2026).

By Kenji Tanaka, Staff Writer