For years, the prevailing assumption in Washington has been that technology is a zero-sum contest, and that the United States can maintain its technological supremacy by building walls around its own innovation ecosystem.

The logic of "decoupling" — whether in semiconductors, artificial intelligence, or cloud infrastructure — is a bipartisan article of faith. But reality has a way of ignoring political wish fulfillment. The reality of 2026 is that Chinese AI models are quietly becoming part of the foundational architecture of cutting-edge applications being built in Silicon Valley.

From Cursor's adoption of Kimi K2.5 to a growing list of US tech companies switching to Chinese foundational models such as MiniMax, Zhipu AI and Qwen, the trend is unmistakable. Eight of the 10 most downloaded open-source models on platforms such as GitHub and Hugging Face are now of Chinese origin. When a premier AI conference, NeurIPS, rejected a Chinese submission, the China Computer Federation's decision to withhold academic services prompted a swift apology — an incident that reveals not just scientific depth but the extent to which global AI discourse now depends on Chinese participation.

What explains this shift? The answer lies in a convergence of developments. The performance gap between Chinese and US foundational models has narrowed dramatically over the past year. Innovations in residual connections and new architectural pathways have placed Chinese models at the frontier of technical capability. The market for AI agents and open-source ecosystems has matured to the point where cost-efficiency and performance matter more than geographic origin. Chinese models have proved exceptionally competitive on both counts. And the global AI industry is beginning to understand that foundational models are infrastructure, much like electricity or the internet. And infrastructure, by its nature, is global. The world's two largest economies cannot afford to simply wall themselves off from each other in AI.

Yet that is what a new bill introduced by US lawmakers seeks to do — banning government use of Chinese robots and, by extension, deepening a "decoupling" agenda that extends to AI. The economic illogic of this approach is staggering. It assumes that supply chains can be untangled overnight, that innovation can be neatly partitioned by nationality, and that the rest of the world will align itself with US restrictions rather than continue to adopt the best and most cost-effective technology available. None of these assumptions hold.

Data is foundational to our digital world. But the creation of infrastructure that produces data and the use of data through AI models have been the focus of attention and not enough meaningful consideration has been given to data itself. The need for a world data body and a comprehensive, multilateral approach for governing the use and flow of data, and to address attempts to hijack the open flow of information by criminals and terrorists, has become increasingly evident. As has the question of who benefits from its expanding uses, as there is a huge gap between the data gains of the developed and developing worlds.

The establishment of the World Data Organization aims to address these issues. The organization held its first general meeting in Beijing on Monday, at which its charter was adopted, its first council elected and the groundwork laid for a multilateral framework to govern data and high-tech development.

As foundational models become general-purpose technologies, their diffusion across borders will determine which economies can participate in the intelligent economy of this century. By championing open-source ecosystems and internationalizing its AI services, China is helping nations bridge the digital divide.

None of this is to suggest that geopolitical competition over technology has disappeared. It has not. But the premise that AI "decoupling" is either feasible or desirable is dangerously mistaken. The industry itself is demonstrating that. When Silicon Valley developers choose Chinese foundational models, they are voting with their code.

What is needed instead is multilateral governance. The World Data Organization offers a framework, however nascent, for precisely that. The alternative — a world of competing technological blocs, fragmented standards, and redundant investment — would impose enormous costs on businesses, consumers and innovation itself. These costs would not be borne equally. They would fall most heavily on economies that are least able to afford them.

Washington's lawmakers would do well to respect reality. The attempt to "decouple" the world's two largest economies in AI is not only impossible; it is a form of self-harm. The infrastructure of the intelligent age will be built on global foundations. The only question is whether the US chooses to help shape that architecture or isolates itself from it.