The geography of global power is being rewritten, not with borders and battleships, but with fiber optics and cooling systems. In the remote plains of Patagonia, the deserts of Chile, and the urban hubs of Brazil, a new kind of race is unfolding, one that will determine not just who controls the world’s data, but who shapes its future. This is the story of the construction of artificial intelligence data centers in Latin America, a narrative deeply intertwined with the concepts of sovereignty, multipolarity, and the quiet, resource-intensive war between the United States and China.

Latin America is rapidly becoming a prime battlefield in the global AI infrastructure race. The region is currently home to over 500 data centers with an installed capacity of nearly 1,450 megawatts . While the U.S. and China vie for dominance, Latin American nations are grappling with a fundamental question: can they move from being mere consumers of AI to becoming its architects?

But to understand what is being built, we must first understand what it is. The structures rising across the region are not the data centers of the past. The difference between a traditional data center and an AI data center is not merely a matter of scale, it is a complete architectural revolution driven by the unique demands of artificial intelligence.

A traditional data center was built for predictability. Its primary workload consisted of general-purpose computing: running applications, hosting databases, and serving websites. These tasks are handled by CPUs (Central Processing Units), which are like versatile project managers, great at handling a wide range of tasks, but performing them sequentially. Standard server racks in these facilities draw a modest 5-15 kW of power, and their cooling is managed by air systems using raised floors and hot/cold aisle containment.

AI changes everything. Training and running large language models require the massive parallel processing power of GPUs (Graphics Processing Units). Imagine not one project manager, but a team of thousands of specialized writers all working on different parts of a document simultaneously. This is what a GPU cluster does, handling the billions of mathematical operations needed for AI.

This fundamental shift in computing creates cascading effects. While a traditional rack consumes 5-15 kW, an AI training rack can easily reach 80-150 kW and is projected to soon exceed 300 kW per rack. Air cooling becomes entirely insufficient at these densities. AI data centers must adopt advanced liquid cooling technologies, either direct-to-chip (DLC) or immersion cooling, to efficiently remove the intense heat from the GPUs. This allows for a much better Power Usage Effectiveness (PUE), the key metric of energy efficiency.

The networking architecture also transforms. Traditional centers optimize for “north-south” traffic, communication between clients and servers. AI data centers require high-bandwidth, ultra-low-latency “east-west” traffic for thousands of GPUs to communicate in parallel, requiring 4-5 times more fiber connections than traditional setups.

Ultimately, the AI data center is not a bigger version of the old model; it is a purpose-built “AI factory” where power delivery, cooling, networking, and compute are engineered as a single, integrated system to support a fundamentally different class of workload.

Latin America is rapidly becoming a prime battlefield in this new AI infrastructure race. The region is currently home to over 500 data centers with an installed capacity of nearly 1,450 megawatts. While the U.S. and China vie for dominance, Latin American nations are grappling with a fundamental question: can they move from being mere consumers of AI to becoming its architects?

According to the Latin American Artificial Intelligence Index (ILIA 2025) , released by the Economic Commission for Latin America and the Caribbean (ECLAC) and Chile’s National Center for Artificial Intelligence (CENIA), the region is a study in contrasts. It accounts for only 1.12% of global AI investment despite representing 6.6% of the world’s GDP. Yet it punches above its weight in adoption, accounting for 14% of global visits to AI solutions and ranking third worldwide in generative AI downloads.

The ILIA 2025 categorizes countries into three maturity levels: Pioneers (Chile, Brazil, and Uruguay), Adopters (Colombia, Ecuador, Costa Rica), and Explorers (more than a third of the region). Among the pioneers, Brazil holds a dominant 37.3% share of the regional data center market, followed by Chile and Mexico at 11.6% each.

The United States views AI dominance not just as an economic advantage but as a national security imperative. The Trump administration’s “Winning the Race: America’s AI Action Plan” frames AI as a zero-sum competition requiring “unquestioned and unchallenged global technological dominance.” This translates into a strategy of exporting “full-stack” American AI technology packages while actively working to exclude Chinese influence.

This approach is palpable in Latin America. In Argentina, U.S. officials have pressured local cooperatives to abandon Chinese technology like Huawei for data centers in strategic regions like Vaca Muerta, one of the world’s largest shale oil reserves. The U.S. Ambassador to Argentina, Peter Lamelas, framed this as an issue of national security: “The energy security of the data is the national security of both countries,” insisting that data must circulate via “networks of confidence” free from Chinese oversight.

The project Stargate Argentina, announced in October 2025, is a prime example. This landmark US$25 billion partnership between OpenAI and the Argentine firm Sur Energy aims to build a massive AI data center in Patagonia. This project, signed under President Javier Milei’s investment incentive scheme, is the first of its kind in Latin America and is designed to put Argentina “at the forefront of the global artificial intelligence ecosystem.” This is a clear U.S. move to secure a foothold in the region’s energy-rich south, a region where the abundant renewable energy and water resources make it ideal for the immense power and cooling demands of AI infrastructure.

The U.S. is fighting an open battle against China’s technological expansion in Latin America. This is most evident in the growing influence of WAICO (World Artificial Intelligence Cooperation Organization), a China-initiated intergovernmental body. In July 2026, Venezuela a country with massive, underutilized energy resources became a founding member of WAICO. The National Review argues this is a direct challenge, stating that “Venezuela’s world-class energy resources should not be under China’s control” and should instead be the “battery of the Americas” for U.S. AI leadership.

China’s counter-strategy is to position AI as a “global public good” requiring multilateral cooperation, offering an attractive path for nations wary of U.S. technological hegemony. This has created a digital dependency trap. While the U.S. offers capital and technology, it often comes with strings attached, demanding the exclusion of Chinese suppliers. The Chinese model, while appearing more cooperative, risks swapping one form of dependence for another.

In the face of this superpower tug-of-war, a powerful movement for AI sovereignty is taking root. This concept, championed by experts and policymakers, argues that nations must control their AI systems to align them with their own strategic, cultural, and security needs.

The most ambitious manifestation of this is Latam-GPT. Coordinated by Chile’s CENIA, this project aims to build a large language model (LLM) with a “Latin American stamp,” trained on over 8 terabytes of text from over 30 regional institutions. The model, with around 50 billion parameters (comparable to GPT-3.5), is built on LlaMA 3 and incorporates not just Spanish and Portuguese but also indigenous languages like Quechua and Guaraní.

This initiative is not just technical; it is deeply political. In November 2025, Peru and Chile signed a memorandum to strengthen cooperation on Latam-GPT, with Peru’s premier stating, “Our region can and should assume a leading role in the debate and in the construction of a digital future.” This collaboration, which now includes Brazil and the Dominican Republic, represents a direct challenge to the idea that Latin America must be a passive consumer of U.S.- or China-made technology.

The prospects for developing a truly sovereign AI are significant but fraught with challenges. On one hand, the ILIA 2025 report highlights that the region’s AI adoption is outpacing its digital weight, and there is clear political will to pursue independence.

However, massive obstacles remain. The investment gap is a chasm: the region receives a minuscule fraction of global AI investment. There is a critical shortage of advanced talent, and a brain drain is accelerating. Most national AI strategies lack funding and implementation mechanisms, making them more aspirational than actionable.

The model being pursued by Latam-GPT, adapting open-source models with regional data is pragmatic but creates hidden dependencies, as the foundational assumptions of the base model remain foreign. As the Brookings analysis notes, “fine-tuning inherits the foundational assumptions of the base model,” meaning a model trained primarily on English text will still carry its original biases.

Perhaps the most compelling argument for Latin America’s role in the global AI race lies not in its code, but in its water and energy. The architectural demands of AI infrastructure, the extreme power density and the need for advanced cooling make these resources critical. An AI data center can consume daily the equivalent of a city of 10,000 to 50,000 people for cooling alone.

The Inter-American Commission on Human Rights has warned that the “accelerated rollout of these facilities…is characterized by intensive water and energy use,” potentially affecting access to water for local communities in Brazil, Chile, Argentina, and elsewhere. In São Paulo, Brazil, which accounts for 58% of regional water consumption in data centers, local communities are already raising concerns about this new form of “extractivism.”

Yet, this resource demand is precisely why the region is so attractive. Argentina’s Patagonia, the site for the new Stargate facility, offers abundant renewable energy. Brazil’s massive hydroelectric capacity is a major draw for companies like Elea Data Centers, which is building a sustainable 30MVA facility for Petrobras that will be powered entirely by certified renewable energy and integrate water reuse systems. The massive Rio AI City project in Rio de Janeiro, planned with a capacity of up to 3.2GW, could be a game-changer.

However, the story is not a simple one of “resource-rich, problem-free.” As the REDESCA report highlights, this resource extraction mirrors past extractive cycles like mining and soy, where the promise of development was often not fulfilled for local populations. South America currently has a fragmented regulatory framework, with Brazil’s WUE (Water Usage Effectiveness) reporting being a rare exception, creating uncertainty about whether these projects will represent genuine technological transfer or a new form of resource colonialism.

The construction of AI data centers in Latin America is a defining moment. The region finds itself at the epicenter of a technological cold war between the U.S. and China, each offering a model of digital development. The U.S. offers capital and dominance, while China offers a potential path away from that dominance, albeit with its own dependencies.

Yet, as analysts argue, this binary may be a false one. Just as the Non-Aligned Movement allowed countries to navigate the Cold War, Latin America may be pursuing a strategy of digital non-alignment. This involves a careful balancing act: using U.S. hyperscalers like AWS and Microsoft for some services, while also incorporating Chinese providers like Huawei and participating in projects like Latam-GPT that are controlled by regional institutions.

The region’s success in the AI era will not be determined by which superpower it chooses. It will depend on its ability to convert its natural advantages, abundant water and energy into genuine technological sovereignty. This means not just hosting the AI data centers those purpose-built “AI factories” with their liquid cooling loops, GPU clusters, and immense power demands but building the talent, the governance frameworks, and the political will to ensure that the AI of the future reflects Latin American values, languages, and realities.

The foundation is being laid, but the true architecture of Latin America’s AI future is still being written. Whether the region becomes a passive host to foreign infrastructure or an active architect of its own digital destiny will depend on the choices made today, in the boardrooms of tech giants, the halls of government, and the communities that will live with the consequences of this new industrial revolution.