Even if it's just to ask a question, create an image, generate a short video, or edit text, consulting an AI chatbot in the cloud activates servers whose operation permanently depends on many physical resources: space, electricity, cooling systems, and primarily,rare minerals and metals, necessary for the manufacture of servers, networks, processing units, sensors, and cabling of data centers that host the service..
The popularization of tools like ChatGPT, Claude, Gemini, and Perplexity in people's daily lives fuels the rapid growth of this infrastructure,which connects the interests of mine and refinery owners to the needs of large data processing complexes..
This expansion opens a new extractive frontier:The economic benefits generated by the new technology circulate through global chains, but a large part of the socio-environmental costs of this exploitation are borne by the countries that supply the raw materials., as few counter-measures. A more just governance of development and the global use of Artificial Intelligence therefore requires identifying and clearly assigning responsibility to those who control it, and who, therefore, should be held accountable for the impacts caused by this new technology.
Industrial scale
Commonly associated with the simplified idea of "the cloud", in fact, the operation of data centers that offer services and operate AI modelsis expensive and harmful to the environmentCurrently, there are approximately12,000 data centersworldwide, including operational units and projects, of which approximately1,360 are considered hyperscale.
In the state of Louisiana, USA, for example, Meta is building a complex that will house theHyperion, a global service dedicated to training AI system developers. The project envisions an area of approximately 930,000 square meters, and announced investments exceedingUS$ 50 billion.
The chain starts in the mines and expands with energy
Servers, chips, and networksMaterials necessary for the operation of AI data centers, large or small, require the use of copper, silicon, gallium, germanium, tantalum, rare earth elements, silver, gold, and platinum group metals. Some of these materials must reach high purity for electronic use. Batteries and power supply systems further increase this dependence.
Manufacturing devices with a strong presence of electronic components can mobilize between50 and 350 timesthe weight of these devices in raw materials.
TheInternational Energy Agencyestimates that existing AI data centers worldwide consumed approximately 485 terawatts (equivalent to 485 trillion watts) per hour, just in 2025. And projects a consumption of 950 terawatts per hour in 2030, considering the entire sector.
Companies announce that a large part of this demand will be met by renewable energy projects. While solar and wind sources are central to decarbonization, their equipment, electricity grids, and storage systems alsorequire large amounts of minerals.
An example of this infrastructure is the solar parkOrion Solar Beltin Texas, which houses over 1.3 million photovoltaic modules of 700 watts each. The project has Google as a key client and provides electricity to the company's data center network.
A hypothetical comparison illustrates the scale: to generate the 950 terawatts per hour projected for 2030, only with 700-watt solar modules, approximately 775 million panels would be needed.
Solar modules can operate for25 to 35 years, but some data center equipment needs to be replaced at much shorter intervals, ofa few years, and this is another major problem. A 2024 study projected the accumulation of1.2 million to 5 million tons of e-waste by AIgenerative AI between 2020 and 2030. Without proper reuse or recovery,Metals are lost and environmental liabilities increaseFor all society.
The economic and social costs in producing areas
In addition, providing minerals does not guarantee any control to the countries from which they are extractedover the technologies with the highest added value generated for the development of AI. As in the 16th century, the tendency is that the gains from the commercialization of the service concentrate in companies, whileThe communities in the producing regions face pollution, occupational hazards, and disputes over land and water.- This is a fundamental element for the long-term operation of large data centers. In this complexity, jobs, taxes, and royalties would need to be accompanied by labor protection, supervision, and public services.
Ourbibliographic and documentary researchgathered evidence of these inequalities. The studied supply chains serve different sectors, including electronics, batteries, and energy systems used by data centers.
According to estimates by theUnited States Geological Survey, the Democratic Republic of the Congo accounted for approximately 73% of global cobalt production in 2025. Recent investigationshave documented the occurrence of child labor, fatal accidents, and exposure to toxic substancesin this activity in the country, which is one of the poorest in the world.
In Indonesia, which is expected to account for approximately two-thirds of the world's nickel production in 2025, research has revealed a processing dependent oncoal, contaminationandterritorial conflicts.
NoChileand inPeru, leaders in copper production, studies have identifiedExposure of communities to toxic metals.
In the Lithium Triangle, formed by Chile, Argentina, and Bolivia, the extraction ofbrinesputs pressure on water resources in arid regions. In China, the mining of rare earths has been associated withwater and soil contaminationIn Myanmar, investigations have documentedpollution and coercionagainst communities in mining areas controlled by armed groups.
The dangerous path between mines and AI companies
To reach the data centers, all these minerals extracted in a harmful way to the environment and society pass throughmerchants, refiners, and manufacturers.
In Rubaya, in the Congo, mines associated with approximately 15% of the global supply oftantalumare under the control of the armed groupM23 in April 2024The metal, obtained from the ore known ascoltan, is used in electronic capacitors. The UN estimated that the group's charges for the mineral amounted to approximately US$800,000 per month, according to data cited byGlobal Witness.
In June 2026, the UN linked the trade of coltan associated with the conflict in the Congo to the Chinese company Ningxia Orient Tantalum, which appears in thesupply chain officially declared by Nvidia, currently the global leader in the production of artificial intelligence chips. Ningxia denied acquiring the material while negotiating with the armed group, but the doubt remains.
In Brazil, aRepórter Brasil investigation, published in 2022, identified connections between illegally mined gold from indigenous lands and refineries present in the declared supply chains of Apple, Google, Microsoft, and Amazon.
The leading companies in the segment themselvesalso recognizethat information from suppliers and audits may be incomplete, inaccurate, or subject to fraud. Publishing lists, therefore, does not equate to correctly verifying the origin of each material.
Water under pressure and confidential information
Evaporative cooling and the electricity required for data centers also consume a lot of water. In the US,researchestimated, in 2023, a direct consumption of 66 billion liters by data centers, and almost 800 billion liters associated with electricity.
The impact of data center operations on the environment varies depending on the climate, the cooling required, and the availability of water. In Brazil, areport published by the Pulitzer Centerin 2025 identified that, of the 22 projects of this type that were being analyzed at the time by the Brazilian government, five were located in municipalities with recurring droughts. In Caucaia, in Ceará, where a project associated with TikTok is planned, there have been emergencies due to drought or drought in 16 years, between 2003 and 2024. The government of Ceará denied access to technical documentation, citing commercial secrecy.
And this secrecy also protects the identity of the companies involved.an investigation by Intercept Brasil, also published on the Pulitzer Center website in March 2026, showed that intermediary companies conduct environmental licensing and electricity connection requests without initially revealing the large technology companies behind the projects.
Theinvestigation into the "AI backyard" in Spain, Chile, and Mexico, also supported by Pulitzer, faced similar obstacles. Spanish authorities denied access to data on water consumption at the facility.Amazon Web Services, which defines itself as the world's most comprehensive cloud of data, claims commercial confidentiality. Without this information, communities and researchers have difficulty assessing the environmental impacts of the project.
Public incentives and corporate responsibility
Governance also involves public resources allocated to attract these projects. In Brazil, theLaw 15.504, of September 2026established the Redata, a tax incentive regime for data centers. Among the provisions are the provision of renewable or low-emission energy, water efficiency, and the publication of sustainability reports for the benefited facilities.
These obligations must translate into verifiable data and oversight. The evaluation of incentives should consider revenues, employment, infrastructure costs, and socio-environmental impacts. In the mineral supply chains, effective governance must require procedures capable of identifying labor abuses and environmental damage, with measures of prevention and remediation.
The infrastructure of AI functions as a new extractive frontier when concentrating benefits and transferring costs to the territories that provide resources. Breaking this dynamic requires linking licenses and incentives to verifiable responsibilities throughout the chain.
Companies that design chips, hire suppliers, or use data centers must be accountable for the conditions that support their businesses. And the communities that live with mines and compete for water with large facilities need to participate in the decisions before the economic interests generated by the investments define their own future, without asking for their permission.
João Stacciarini received a postdoctoral funding from the Coordination for the Improvement of Higher Education Personnel (CAPES).
Ricardo Assis Gonçalves receives funding from the National Council for Scientific and Technological Development (CNPq).