How AI firms are appropriating Latin America’s resources

In October 2025, OpenAI – the US company behind AI chatbot ChatGPT – and Argentinian firm Sur Energy jointly announced their intention to spend $ 25 billion building a new data centre in Argentina. The plan is to invest in Patagonia, a region in the south of the country boasting a great variety of nature reserves. It’s not yet clear exactly where the project will be sited. It didn’t take the Argentinian government led by Javier Milei long to enthusiastically promise that the project would transform the country into a regional and global hub for artificial intelligence. The deal was quickly sealed. 

Several Latin American states are keen to jump on the “new digital era” bandwagon and become AI pioneers. Besides Argentina, they include Brazil, Mexico and Chile. Thanks to their size and resource wealth, these countries dominate the sector in the region. They ensure the industry has sufficient energy and water – even as local populations repeatedly suffer water shortages and power outages. This blatant contrast has been highlighted and criticised time and again by regional media such as the transnational journalistic research project “Big Tech’s Invisible Hand”, which is coordinated by the Brazilian platform Agência Pública and the Latin American Centre for Investigative Journalism (Centro Latinoamericano de Investigación Periodística, CLIP). All the same, there’s a distinct absence of any broad-based public debate about the construction of new AI infrastructure in Latin America.

As global demand for AI applications soars, data centres also need ever more computing capacity and energy to operate. Asking questions of generative AI models such as ChatGPT is known to consume far more energy than using a search engine like Google. To meet this demand, what are known as hyperscale data centres are necessary. Based on complex and costly infrastructure, they use significantly more power in total than their predecessors. According to the International Energy Agency (IEA), the industry is currently responsible for one to two percent of global electricity consumption – a figure expected to rise to three percent by 2030.

The US government and companies based there are perfectly well aware that they will need far more energy to expand their data centres. They see Latin America as a game-changer. Speaking before the US Senate Committee on Foreign Relations in May 2025, US Secretary of State Marco Rubio for example called for a debate about which countries could help meet the demand for AI energy. Rubio stressed that energy would be at the forefront of American foreign policy “for the next 100 years”. “We need to discuss how we will invest in countries with the kind of energy supply that can meet that demand,” he said. He cited Paraguay as an example, claiming it had surplus energy capacity from hydroelectric power following the expiry of its contract with Brazil.

Expanding power grids

Energy, be it from renewable or fossil sources, is crucial to the AI industry. In response to the rampant growth of data centres in Latin America, some countries have decided to expand the capacity of their power grids. In 2024, Mexico’s Federal Electricity Commission (CFE) announced that the capacity of the electricity network in the country’s Querétaro state would be stepped up by 50 %, citing the construction of data centres in the region as one of the main reasons. The CFE also declared that a new gas-fired power station would be built to produce electricity. More electricity generated using gas means more climate-damaging emissions. 

Alongside energy, water has traditionally been a vital resource in data centre infrastructures. Used to cool buildings, it keeps temperatures at a steady 18 °C to 27 °C to prevent servers from overheating. This evaporative cooling method with water is no longer used these days, at least not by OpenAI, stresses the firm’s CEO, Sam Altman. However, though new cooling systems promise to slash water consumption, or indeed eliminate it completely, it will be a long time before they are deployed in all data centres. Altman was thus unable to refute allegations that the industry continues to use substantial quantities of water.  

It is simply a fact that some data centres are being built in specially protected regions or areas already suffering from water stress. Neuquén province in Patagonia, for example, where OpenAI is planning its data centre, is facing increasing water scarcity. And the AI industry also has its sights on Querétaro, despite the veritable water crisis the state is facing. It knows that governments tend to show political flexibility when it’s a question of attracting AI investments. Indeed, by 2025 the Sustainable Development Ministry (Secretaría de Desarrollo Sustentable) in Querétaro had already greenlighted 20 projects for additional data centre infrastructure – even though a local environmental agency had recommended not granting any new approvals involving water consumption, as the “Big Tech’s Invisible Hand” project discovered. 

One law for the industry

This raises an important question: What limitations should countries in Latin America and elsewhere impose on the AI industry? As things currently stand, regulations are often flexible or don’t exist at all. In fact, the AI industry engages in massive lobbying to ensure that laws are tailored to suit it. Government representatives meet with senior executives of companies like Microsoft, Amazon, OpenAI or Google, and countries sign deals worth millions with them. The governments know full well that their countries will have to offer something if they want to be part of the new AI era: building land, inexpensive energy and other resources, industry-friendly tax incentives or simply agreements that allow the tech giants to continue expanding their operations.

The narrative governments tell their citizens is that the investments will boost the local economy, arguing that new jobs and development will be the result. At the same time, however, social organisations are increasingly critical of big tech companies – because of the environmental impacts and other shortcomings in terms of regulations. They rightly ask key questions such as: What long-term effects will the AI data centres actually have for the local population and the natural environment it relies on for its livelihoods? Who will have control over the data? And could AI perhaps be used in a different, more sustainable way? Governments would be well advised to listen to these critical voices to make sure that AI investments don’t end up doing more harm than good. 

Julia Gavarrete is an investigative journalist from El Salvador.
Bluesky: @PetizaGavarrete

This story is part of The 89 Percent Project, an initiative of the global journalism collaboration Covering Climate Now.