In short
Beijing is following the same playbook it used for solar panels and electric vehicles: flooding the world with cheap, open-weight models and setting industry standards. While the U.S. tightens export controls, the Global South is opting for Chinese models—even at the cost of dependence.
Leaders from American frontier labs did not attend the AI for Good conference in Geneva—Mark Benioff turned out to be the most prominent CEO there. In contrast, the Chinese delegation was massive and spent four days convincing ministers from Pakistan, Zambia, Russia, and the entire Global South that free Chinese models are their future. Wang Jian, a former top executive at Microsoft Asia and the architect of Alibaba’s cloud business, put it bluntly: Chinese AI can be a “resource” for other countries, just like energy. “It’s very important for the rest of the world to have a choice.”
This isn’t charity. It’s the same strategy China used to capture the solar panel and electric vehicle markets: first, flood the world with cheap products; then, set the standards; then, control the infrastructure. Only now, instead of ports and railroads, it’s tokens. A week after Geneva, Xi Jinping declared at the World AI Conference that we must “seize this rare historical opportunity” and promote open source, adding that “AI development should not be a solo performance by a single country.”
The U.S. is responding with export controls and selective cooperation limited to “trusted allies.” The Treasury Department has explicitly called the reverse engineering of American models by Chinese laboratories intellectual property theft. The problem is that drastic export restrictions—such as the recent ban on the export of Anthropic models—have already caused serious disruptions for non-U.S. companies and governments. The tighter Washington turns off the tap, the more attractive China’s “AI without borders” proposal looks.
For practitioners—developers, government agencies, and companies in developing countries—the choice isn’t just about benchmarks. Chinese open-weight models are free and accessible, but their training data is not disclosed, and research shows a built-in Beijing-centric perspective on sensitive topics. A red team that tested the DeepSeek assistant for an economic agency found China’s stance on Taiwan’s semiconductor industry reflected in the responses. On the other hand, open-weight models at least allow users to customize the model to their needs, something that closed American counterparts do not offer.
The director of Cameroon’s IT agency explained that the pilot model for civil servants runs on an IBM mainframe, not on a Chinese system. Tanzania, whose fiber-optic network was built by the Chinese using Chinese loans, is developing a sovereign model that can be built on top of either Chinese or American infrastructure. This is the reality of the market: no one wants to trade dependence on one superpower for dependence on another.
China is currently playing the role of Android against the American Apple—more open, cheaper, and more accessible. But symbolic victories are not the same as substantive ones: as analyst Kyle Chan notes, Chinese AI diplomacy still lacks “much substance.” The question that will determine the outcome is whether Beijing can resolve the issue of trust and transparency regarding training data. So far, there is no answer—the free token remains free, but its consequences are far from cheap.