The language of an AI race suggests a sprint between two or three giants, but the most consequential politics may unfold in the space between them, where Latin American countries and other middle powers are forced to navigate overlapping technological, economic, and security dependencies.
Instead of choosing whether to build the next leading frontier model, governments in Brazil, Pretoria, or Jakarta face a subtler problem: how to avoid becoming mere data mines, testing grounds, or surveillance customers for someone else’s AI strategy. The risk is not just technological dependence; it is a gradual erosion of policy autonomy as core state functions are rewired around foreign-owned infrastructures, standards, and platforms.
At the heart of this emerging order remains the rivalry between the United States and China, where AI has become a central theater of great power competition: Beijing treats AI as a pillar of comprehensive national power, embedding it in industrial policy, military modernization, and the export of security technologies, from facial recognition and smart city platforms to data management tools sold across Africa, Asia, and Latin America.
Washington has in turn fused industrial strategy with national security, tightening export controls on advanced chips and manufacturing equipment, subsidizing domestic semiconductor and cloud infrastructures, and pressuring allies to align on controls and standards. The result is a form of digital mercantilism: algorithms, data centers, and semiconductor supply chains are treated as strategic assets, and access to them becomes a lever of diplomatic pressure and coercion.
Focusing only on this binary rivalry obscures how the AI race is also a story about sovereign AI and differentiated state-building among middle powers. Countries such as Brazil, Mexico, India, and South Africa are experimenting with their own AI strategies, not to “win” in model capabilities, but to avoid technological vassalage to either Silicon Valley or Shenzhen. They are investing in domestic data infrastructures, language models trained on local languages and legal systems, and regulatory regimes that embed local norms about privacy, labor, and content. In doing so, they turn AI into a new instrument of state-building: ministries are reorganized, new digital agencies are created, and public-private consortia emerge to manage cloud resources, national datasets, and critical applications in areas like health, taxation, and social policy.
This competition already has a hard security edge: AI-enabled systems are changing the conduct of conflict and internal security, from automated image analysis and cyber defense to semi-autonomous drones and predictive policing. Militaries experiment with swarming systems, target recognition, and AI-assisted intelligence analysis that promise speed and efficiency but also increase opacity and compress decision time in crises, which can heighten the risk of miscalculation. Unlike nuclear systems, many AI capabilities are dual-use and software-based, making them easy to diffuse and hard to monitor. In fragile regions, including parts of Latin America where security forces already struggle with accountability, the quiet delegation of judgment to opaque systems could entrench abuses under a veneer of algorithmic objectivity.
Economically, AI is catalyzing a massive reallocation of capital and attention that binds it to energy and infrastructure geopolitics. The scramble for high-performance chips, hyperscale data centers, and the electricity to power them ties AI to questions of grid capacity, water use, and climate policy. Countries that can combine reliable energy, advanced manufacturing, and hospitable regulatory environments will become hubs for AI-intensive industries.
For many middle-income states, the danger is that they are locked into lower-value roles: hosting data centers without capturing governance power, providing labeled data through low-paid digital labor, or adopting imported AI applications that displace local firms without building domestic capabilities. This pattern risks reinforcing core–periphery dynamics in the digital economy.
It is in the export of AI-enabled surveillance and governance tools that the asymmetry is most visible: Chinese companies, often backed by state financing, market safe city packages, facial recognition systems, and integrated command centers to Latin American governments seeking to tackle crime and social unrest. Western firms provide alternative offerings through cloud-based services, content platforms, and analytics tools tied to their own standards on privacy and content moderation.
These systems travel with embedded assumptions about state-society relations and acceptable levels of opacity: a municipality in Latin America that adopts a turnkey surveillance platform from abroad may find that its policing priorities, its categories of “risk,” and even its data retention practices are effectively pre-configured by foreign engineers and vendors.
Consider a concrete Latin American scenario: under pressure to respond to public insecurity, a government signs a generous credit-backed deal with a Chinese or Western consortium to deploy AI-driven cameras, facial recognition, and predictive policing software in major cities. In the short term, authorities present this as a modernization success: crime heat maps, dashboards, and rapid response centers promise a more efficient state.
Over time the city’s policing strategy and resource allocation become tied to proprietary algorithms trained on local data but controlled by a foreign vendor. When civil society groups demand transparency on false positives, discrimination, or political surveillance, officials are told that the code is a trade secret. The country finds itself in a position where challenging the system means jeopardizing financing, technical support, and access to future updates, constraining its democratic oversight.
Another example lies in the management of critical infrastructure such as energy, logistics, and agriculture. A Latin American country might adopt AI platforms from large foreign cloud providers to optimize its power grid, ports, or soy supply chain. These systems can deliver efficiency gains and emission reductions, but they also centralize data and decision support in infrastructures governed under foreign jurisdictions and corporate policies. In a period of diplomatic tension, sanctions, or corporate disputes, access to updates or services could be restricted, affecting everything from port logistics to food exports. The AI race, in this sense, creates new choke points: not only in chips and hardware but also in the embedded AI services that make complex systems intelligible and manageable.
There is also a softer, cultural dimension: Generative AI models trained predominantly on English-language content, Western legal frameworks, and particular political narratives risk reproducing biases when applied to Latin American contexts. If local media, education systems, or public communication increasingly rely on such models, subtle shifts in language, framing, and historical interpretation may occur.
For instance, automated translation tools or content-generation systems might systematically underrepresent Indigenous languages and perspectives or render regional political debates through categories imported from elsewhere. This is not censorship in the traditional sense, but it is a gradual shaping of the informational environment that can affect public discourse and policy imagination.
Middle powers are not passive in this story, and some are actively trying to exploit the AI race to their advantage. A country like Brazil could pursue a strategy of becoming a regional hub for responsible AI, investing in open models trained on Portuguese and regional data, aligned with local legal standards and deployed through public cloud infrastructures with strong oversight.
It could leverage its diplomatic weight to convene coalitions around AI in agriculture, climate adaptation, or health, positioning itself as a bridge between Global North technology producers and Global South users. Similarly, Mexico might choose to integrate AI into its manufacturing and logistics sectors in ways that deepen its role in North American supply chains while insisting on joint governance arrangements for critical data and models.
Outside Latin America, countries such as India, Indonesia, or Nigeria face analogous choices. They can accept a future where their AI ecosystems are effectively extensions of American or Chinese platforms, or they can invest in domestic institutions and standards that give them room to negotiate. This could involve mandating algorithmic transparency for public-sector applications, requiring local data storage for certain sensitive domains, or supporting indigenous and local language technologies through public procurement.
It might also mean aligning in flexible, issue-based coalitions rather than rigid blocs, for example, collaborating with European actors on AI regulation while engaging with U.S. and Asian firms on industrial applications.
If this is the landscape, the metaphor of a race becomes both descriptive and dangerous: it captures the intensity of investment and the sense of urgency among elites in Washington, Beijing, and Brussels, who fear that falling behind in AI equates to a loss of geopolitical standing. Yet races imply a finish line and a single winner, while AI is a general-purpose technology whose impacts will unfold unevenly across decades and domains.
The race frame encourages secrecy, zero-sum thinking, and the sidelining of common risks, from accidents in autonomous systems to the concentration of economic and informational power in a handful of corporations and security agencies. It also makes it harder for leaders in middle-income countries to argue for a more measured, governance-first approach without appearing naïve or defeatist.
A more realistic geopolitical narrative would treat AI as a shared, contested infrastructure rather than a trophy. This would involve seeing export controls on chips not only as tools to outcompete rivals but also as instruments that can lock whole regions into particular technological spheres or push them towards costly attempts at autarky. It would require acknowledging that while U.S. and Chinese AI ecosystems are increasingly politicized, they remain intertwined through research networks, supply chains, and global markets, creating both vulnerabilities and openings for negotiated limits.
Crucially, it would mean taking seriously the agency of middle powers and regional actors that are using the language of sovereign AI to carve out space for experimentation with governance models, safety standards, and public-interest applications.
For Latin America and other regions sandwiched in this contest, the main strategic question is not whether they will “win” or “lose” the AI race as measured by benchmark scores in frontier labs. It is whether they can use AI to expand their developmental and democratic horizons without becoming locked into dependent relationships with external providers of infrastructure, standards, and security.
That will require a politics that goes beyond procurement decisions and technical pilots, towards deliberate choices about which public functions should be digitized, which should remain under direct human and institutional control, and how to ensure that imported systems remain legible and accountable to local societies. The alternative is a future in which AI deepens the structural imbalances of the international system, leaving those in the middle running fast but on someone else’s track.
