security, and global influence.2 This dynamic will have significant repercussions for whether and how developing countries can access AI. Incentives and asymmetric dependence shape technology choices Since 2018, the United States has introduced export controls on advanced semiconductor chips and related supply chains, with exports to restricted entities requiring approval by the government.3 Meanwhile, China has leveraged its control over critical raw materials by announcing a ban on the export of technologies to process rare earths, as well as export controls on the rare earth elements themselves.4 China’s latest measures also target technological know-how and foreign products made with Chinese inputs. However, several of the latest measures have been suspended until November 2026.5 Why do the major powers promote the use of their AI technology? The first reason relates to scale. As discussed in part 1, AI benefits from economies of scale at many layers of the stack. The second reason involves influence. Having control over the AI infrastructure determines whose rules become the dominant global standard. More than 80 percent of the world’s population lives in developing countries. Every additional country whose infrastructure, developers, and users are tied to a given technological ecosystem contributes to scale economies, strengthening the influence of the leading powers. Key layers of the AI stack can function as strategic “choke points”: Control at each layer constrains who can participate in the AI value chain and acquire the capabilities of AI. Although open-source and open-weight models provide developing countries with access to models,6 running models at scale requires either domestic data centers, which would have to be built on chips controlled by major powers, or commercial cloud providers based in some of these countries. Countries may seek to reduce dependence in several ways. One approach is to source different layers of the AI stack from different countries, choosing the best available technology for each layer. However, because AI systems depend on these layers working together, this strategy can create interoperability problems and expose countries to conflicting regulations tied to technologies from different jurisdictions. Pursuing “AI sovereignty” is an alternative approach, but as will be discussed, it is often a costly and ineffective one. Part 1 argued that AI offers developing countries an opportunity to ease shortfalls in skilled capabilities. Mishandling such trade-offs could cost developing economies access to this very technology. Notably, these powers often use incentives to promote their technology through financing and market access. The American AI Exports Program, launched under a US executive order in July 2025, promotes exporting US-developed “full-stack” AI technologies consisting of hardware, software, and services to allies to secure global technology leadership.7 China’s Digital Silk Road, launched in 2015 as part of the Belt and Road Initiative, aims to build crucial infrastructure in developing countries, including 5G networks, undersea cables, AI laboratories, data centers, and smart cities.8 Beyond incentives, asymmetric dependencies can also influence a country’s technology choices. One common form of asymmetric dependency is that of trade.9 Developing countries often rely heavily on a single major economy for trade (refer to figure 6.1). For example, developing countries in Latin America export a disproportionate share of their goods to the United States compared with other major powers, while those in East Asia and Pacific export an approximately equal share of goods to China and the United States—suggesting the presence of asymmetric dependencies with not just one but both countries. AI’s Political Impact: Reshaping Power Within and Across Countries 203