once accessed globally. By contrast, incumbent firms from the major powers continue to benefit by providing the AI stack and selling to the fragmented markets. For developing countries, neither overdependence nor excessive fragmentation offers a sustainable path. Most countries cannot afford to build every layer from scratch. Thus, for most countries, the middle path between sovereignty and dependence is a more viable strategy. This entails, where possible, diversifying suppliers within each layer to reduce the risk of choke points while ensuring interoperability. It also involves targeted steps to reduce dependence, such as using open-source and open-weight models, establishing strong data governance, engaging in standards setting, and forming coalitions with like-minded countries to boost collective bargaining power. Part 3 of the Report builds upon such initiatives. This approach accepts that there will be trade-offs. Some choke points will persist, but this approach maintains access to economies of scale and frontier performance that outright sovereignty strategies would forgo. Dynamics between governments and corporations: Entanglements could impede development Within a country, the relationship between ­governments and corporations affects whether the gains from AI are broad based, wasted, or concentrated in the hands of existing power holders. Governments can lean too far toward accommodating firms, offering preferential terms to attract investment, at the expense of public funds. At times, governments may exercise their regulatory and enforcement powers in an unanticipated way, creating uncertainty for firms. In some instances, the relationship can settle into arrangements that are too cooperative, entrenching the interests of both through government-backed contracts flowing to ­ connected firms and firms aiding in the extension of state power over citizens. Each of these phenomena can be partly attributed to the same underlying cause of many imbalances of power between firms and governments. Firms lobby at scale.22 The number of firms lobbying in the United States on AI-related issues nearly doubled between 2019 and 2025,23 even as lobbying activity concerning issues such as immigration remained relatively constant during this period (refer to figure 6.3).24 Active industry engagement in the European Union led to a shift in regulatory approach toward innovation and competitiveness.25 An extensive investigation by the Centro Latinoamericano de Investigación Periodística and various partners documented nearly 3,000 lobbying actions across multiple economies between 2012 and 2025, with such activity rising rapidly in Latin America.26 Even without lobbying, these companies wield significant influence stemming from their position across the AI stack.27 Moreover, because throughout the AI ecosystem switching costs are substantial, the possibility that a firm may reduce operations or withdraw owing to unfavorable regulations is credible, even without explicit threats. In some instances, because the technology is evolving so quickly, as government officials and agencies formulate policies and regulations, they may rely on the very corporations that they are tasked with regulating to explain the technology’s abilities and potential risks. Governments are not powerless. They retain significant levers such as export controls,28 regulating access to domestic markets,29 designations of supply chain risk,30 and procurement rules (refer to chapter 8), as well as acquiring stakes in companies.31 However, it is not easy to strike the right balance in using these levers. When used too readily, they can create uncertainty for firms, but when used too sparingly, governments may become too accommodative to firms or settle into arrangements that are too cooperative. AI’s Political Impact: Reshaping Power Within and Across Countries 207