SHANGHAI, China — China has formally established a new international organization for artificial-intelligence cooperation, bringing together 29 countries in an effort that could expand Beijing’s influence over the rules, technologies and infrastructure underpinning the global A.I. economy.

Representatives of the founding countries signed an agreement in Shanghai on Thursday establishing the World Artificial Intelligence Cooperation Organization, or WAICO. According to the agreement, WAICO will be an independent intergovernmental organization headquartered in Shanghai. Its founding members include Russia, Brazil, Indonesia, Pakistan, Kazakhstan and Laos. No European Union country, the United States, Japan or South Korea was among the initial signatories.

President Xi Jinping personally addressed the World Artificial Intelligence Conference the following day, presenting China as an advocate of broadly accessible technology and a more inclusive international system of A.I. governance.

Mr. Xi called for stronger human oversight, legislation, risk monitoring and safeguards to ensure that artificial intelligence remained “safe and controllable.” At the same time, he urged countries to resist technological monopolies and restrictions that could prevent developing economies from gaining access to advanced systems.

The message contained an unmistakable geopolitical challenge: China wants to become not only a producer of competitive A.I. models, but also an architect of the institutions governing their use.

A New Institution, but Not an ‘A.I. United Nations’

Chinese officials describe WAICO as a response to demands from the Global South for greater representation in decisions about artificial intelligence. The organization says it will promote development that is beneficial, safe and equitable, following the principles of the United Nations Charter. U.N. Secretary-General António Guterres attended the signing ceremony, according to China’s Foreign Ministry.

But describing WAICO as an “A.I. United Nations,” as some commentaries have done, goes too far. It is not part of the United Nations system, and its authority, financing, decision-making procedures and practical influence have yet to be demonstrated.

Its founding membership nevertheless gives the organization political significance. By assembling countries from Asia, Africa, Latin America and parts of Europe outside the European Union, Beijing has created a permanent forum through which it can promote its preferred standards for A.I. development, security and regulation.

China’s official account emphasizes multilateral cooperation rather than confrontation. Western analysts, however, see the organization as a potential counterpart to American-led initiatives such as Pax Silica, which focuses on trusted technology and semiconductor supply chains. The European Union, Germany and several other European countries recently joined that American-backed framework.

WAICO should therefore be understood less as an explicitly anti-Western alliance than as part of a developing competition between overlapping technological blocs.

China’s Open-Model Diplomacy

China’s strongest advantage may not be the new institution itself, but the technology and pricing strategy that accompany it.

Chinese companies including Alibaba, DeepSeek, Zhipu AI and Moonshot AI have released models whose weights can be downloaded or deployed independently. Their services are frequently cheaper than comparable American offerings, making them attractive to businesses and governments with limited computing budgets.

At the Shanghai conference, Moonshot introduced Kimi K3, a 2.8-trillion-parameter, open-weight model that the company says can compete with leading American systems in coding, reasoning and autonomous tasks. It is considered one of the largest models of its category, though parameter counts alone do not establish performance: Kimi K3 uses a mixture-of-experts architecture, meaning only part of the model is activated for each request.

Calling China’s approach “free A.I. for the world” is also misleading. Open weights do not necessarily provide access to training data, source code or the full development process. Operating such enormous systems still requires substantial computing infrastructure, and commercial services may carry usage fees or licensing conditions.

Nevertheless, China’s approach offers developing countries something tangible: affordable models, technical training, deployment support and fewer immediate export restrictions. Mr. Xi announced 5,000 training opportunities for developing countries over five years and additional technology programs, including A.I.-based meteorological services for 30 countries.

American Restrictions Strengthen China’s Argument

China’s diplomatic campaign has gained credibility from recent American restrictions.

In June, the United States directed Anthropic to prevent foreign nationals from accessing its Fable 5 and Mythos 5 models, citing national-security concerns. Anthropic subsequently disabled the two systems for all customers because it said it could not reliably separate users by nationality. Its other models remained available.

The episode did not amount to a general American closure of A.I. technology, as some reports have suggested. But it raised doubts among international companies about their dependence on proprietary American systems that could be withdrawn under government orders.

Some European and American companies, including Siemens, Airbnb and DoorDash, have begun using Chinese models alongside Western ones, according to the Financial Times. Cost is an important factor: certain Chinese services can be dozens of times cheaper, while downloadable weights allow companies to run models on their own infrastructure.

OpenRouter data also indicate that Chinese models have overtaken American competitors in token consumption on that particular platform. That is evidence of growing usage, not proof that China has surpassed the United States in total computing capacity or technological efficiency.

The Technological Gap Is Narrowing

China’s technical progress is real. Stanford University’s 2026 AI Index concluded that the measured performance gap between leading American and Chinese models had “effectively closed.” As of March, the leading American model retained an advantage of about 2.7 percent in the report’s aggregate comparison, and models from the two countries had repeatedly exchanged positions near the top of benchmark rankings.

That finding does not mean the two national A.I. industries are equal in every respect. The United States continues to produce more prominent frontier models and benefits from greater private investment, advanced semiconductor access and the dominance of American cloud providers. China leads in the volume of A.I. publications, citations and patents and has demonstrated an ability to build competitive models despite American chip-export controls.

Questions also remain about how some Chinese developers achieved their rapid progress. Anthropic has accused operators associated with Alibaba’s Qwen laboratory of using nearly 25,000 fraudulent accounts and more than 28.8 million interactions to extract capabilities from Claude through unauthorized model distillation.

Alibaba has been accused, not convicted, and distillation itself is a widely used and legitimate training method when conducted with permission. The unresolved issue is whether restricted model outputs were obtained deceptively and used in violation of contractual or intellectual-property protections.

Europe Is Behind, but It Is Not Standing Still

The most exaggerated part of the original argument is that Europe has no competitive A.I. industry and is doing little beyond debating regulation.

Europe unquestionably trails the United States in frontier-model development, venture financing and cloud infrastructure. It is also dependent on American and Asian semiconductor suppliers. Its market remains fragmented, and critics contend that regulatory complexity makes it harder for European companies to expand.

But Europe is not without capable companies, research institutions or an industrial strategy. It has developers such as Mistral AI, strong scientific institutions, advanced supercomputing facilities and valuable industrial data in manufacturing, energy, automotive engineering and pharmaceuticals.

The European Commission’s AI Continent Action Plan calls for mobilizing €200 billion in investment, creating at least 19 A.I. factories and financing as many as five A.I. gigafactories. Its Apply AI Strategy also promotes European and open-source systems in strategic industries and public procurement.

Whether those plans will be implemented quickly enough is a legitimate question. But reducing Europe’s response to the AI Act alone presents an incomplete picture.

A Contest Over Dependence, Not Just Performance

The emerging divide is more complicated than a choice between cheap Chinese models and expensive but trustworthy American ones.

Chinese systems may offer lower prices and greater deployment freedom, but they raise questions about data security, censorship, state access and long-term technological dependence. American models often benefit from stronger global commercial ecosystems, but export controls and unilateral access decisions create their own sovereignty risks.

Europe’s challenge is therefore not simply to choose one side. It is to develop enough domestic computing capacity, models and technical expertise to avoid becoming permanently dependent on either.

WAICO will not immediately determine global A.I. rules. Its importance lies in showing that China now possesses a coordinated alternative: competitive models, affordable access, institutional diplomacy and a receptive audience across much of the developing world.

The global A.I. order is not yet divided by an iron curtain. But the foundations of rival technological systems are becoming increasingly visible — and the countries that fail to build their own capabilities may eventually find that the most important rules have been written elsewhere.