This is a provisional English translation of a report originally published in Portuguese by IP.rec. A revised and fully formatted edition will be released in the coming months.

Authorship: André Lucas Fernandes, Carolina Gomes Pugliesi Branco, Clarissa Mendes Gonçalves, Luana Batista Araújo, Raquel Lima Saraiva.

Introduction

“We are living through a disruptive moment.” “A major window of opportunity.” “The new industrial revolution.” One of those “historical milestones that change the course of humanity.” This was the tone adopted in several of the speeches delivered at the opening session of the Special Committee on Artificial Intelligence of the Brazilian Chamber of Deputies on May 20 of this year. The central concern was whether Brazil would be able to seize the moment and embark on the so-called path of progress.

The field of Artificial Intelligence has indeed risen sharply in recent years, driven by exponential growth in computational capacity and real technological advances, even though speculation and marketing also play a significant role.

The lack of clarity surrounding the environmental impacts of artificial intelligence begins with its very name. Although the term evokes abstract and immaterial dimensions, none of it functions without the natural and mineral resources required to build its components. In Atlas of AI, Kate Crawford argues that computational technologies now actively participate in the geological and climatological processes of the planet. Non-renewable elements that took billions of years to form within the Earth are extracted, used, and discarded at high speed for the immediate advancement and use of contemporary technologies. As Crawford notes:

“[…] this extraction to sustain the technology sector is carried out while keeping the real costs out of sight. Ignorance about supply chains is built into capitalism, from the way companies shield themselves through contractors and subcontractors to the way products are marketed and advertised to consumers.” (Crawford, 2021: 31–32)

Against this background, this work seeks to explore the environmental opacity that permeates the AI lifecycle and the mechanisms mobilized to create the impression that AI is a clean technology with no significant impacts. We examine how the AI industry fits within an extractivist developmental paradigm in which resources are exploited and consumed in predatory ways, often without adequate ethical or environmental considerations. The broader objective is to provide a critical contribution to thinking about models of innovation that do not merely reproduce established formulas.

1. Environmental Opacity in AI

Opacity, in varying degrees, can be understood as partial or total ignorance about a given subject, particularly associated with a lack of transparency in processes. This ignorance may be maintained through several mechanisms: negligence, whether deliberate or inadvertent; secrecy or suppression of information; destruction of documents; or the dissemination of ambiguous, contradictory, or misleading information that reinforces doubt, denies responsibility, or diverts attention from a given issue (Proctor, 2008; Burke, 2024).

In The Triumph of Doubt, David Michaels (2024) examines the strategy of manufacturing uncertainty regarding evidence that certain products used or produced by corporations are harmful to public health. Instead of adopting precaution, these strategies demand definitive proof of harm before action is taken. In addition to avoiding responsibility for damages, corporations seek to gain time and delay or prevent regulations that would limit their operations. The tactics range from data manipulation to commissioning research designed to produce favorable outcomes. The tobacco industry remains one of the most emblematic cases of such practices.

Within AI literature, similar concepts have been used to describe comparable dynamics. Kate Crawford refers to a strategic amnesia surrounding technological progress, particularly regarding its long history of devastation, such as in mining (Crawford, 2021: 26). Tarcízio Silva discusses strategic disinformation, especially by private sector actors and neoliberal political movements, to prevent negative technological impacts from becoming publicly visible (Silva, 2022: 217). Both emphasize that the absence of information is not accidental.

There are several reasons why digital technologies are perceived as clean. The first is conceptual. The term “artificial intelligence” suggests abstraction and immateriality. Data are stored in “clouds,” an expression that implies something ethereal. Technology companies frequently reinforce this association. Claims such as “Digitalization reduces material costs,” “The green transition is only possible through digital technologies,” “Blockchain can help monitor and measure carbon emissions,” and “AI technologies are the solution to the climate crisis” contribute to this perception (Fritsch and Van der Waal, 2022).

What remains concealed is the physical and human infrastructure required for AI to exist. AI systems demand rare earth minerals, oil, coal, large quantities of water and energy, and precarious human labor. In software terms, the development of natural language processing and computer vision models requires massive amounts of energy, a demand that has increased with generative models. This extractive process is reshaping the Earth, yet its costs are rarely disclosed by the industry, whose projects are largely oriented toward attention capture and hype generation rather than environmental problem solving (Crawford, 2021: 15).

Mineral extraction forms the backbone of the AI economy. It depends on lithium, tin, gold, tantalum, and tungsten. Lithium, for example, is essential for rechargeable batteries. Many of these materials must be excavated and chemically processed, generating deep environmental impacts and often accompanied by local and geopolitical violence. As Crawford argues, mining remains highly profitable for those who promote it precisely because accountability mechanisms for its real costs are weak or absent, leaving local populations to bear the consequences.

In Brazil, this dynamic is visible in the Amazon, where illegal mining operations have heavily affected Indigenous territories and contributed directly to the humanitarian crisis faced by the Yanomami people. Illegal mining drives deforestation, contaminates rivers with mercury, and attracts organized crime (Camargos, 2022). A similar pattern can be observed in the Democratic Republic of the Congo, the world’s leading cobalt producer, where cobalt extraction underpins high-tech industries amid a severe humanitarian crisis (Montanini, 2024).

Such dynamics are enabled by business structures that rely on layered contracting and subcontracting, obscuring supply chains and making lines of responsibility difficult to trace. In 2024, Reuters reported that in the Congo case, third-party audits and certifications were used to shield mineral supply chains from scrutiny (Rolley, 2024). Comparable practices have been documented in gold extraction in the Bolivian Amazon (Radwin, 2022).

The materialization of the “cloud” becomes evident in data centers, which underpin services such as social media, email, and banking applications. Data centers are among the largest consumers of electricity globally (Teixeira, 2024) and account for 45 percent of greenhouse gas emissions in the technology sector (Vianna, 2022). When located in regions where energy is generated from fossil fuels such as oil or coal, their impact increases (Ferrari, 2023). Water consumption is also substantial. In 2021, Google consumed 17.4 billion liters of water, primarily for cooling data centers. Meta consumed 2.5 billion liters. Apple, Tesla, and Shell have not provided detailed public information on water use (Williams, 2023).

These figures may underestimate the actual impact. According to the Uptime Institute’s Global Data Center Survey (2024), most data center operators do not monitor water use. Only one third measure carbon emissions and electronic waste disposal. An investigation by The Guardian estimated that the real carbon emissions of data centers operated by Google, Microsoft, Meta, and Apple may be approximately 662 percent higher than officially reported (O’Brien, 2024). One strategy that contributes to this opacity is the use of unbundled renewable energy certificates, which allow companies to claim renewable energy usage even when electricity is sourced from fossil fuel plants (Rathi and White, 2024; Peña, 2025).

The issue extends beyond data centers. A faithful assessment of these systems requires analysis of the entire production chain and the global lifecycle of highly complex infrastructures. Transparency is further constrained by trade secrets, intellectual property protections, and logistical and technical complexity. Absolute transparency may neither be possible nor necessary. The following sections address these limits.

2. Regional Dimensions of Environmental Impacts

Beyond treating opacity as mere ignorance, it is necessary to recognize its structural dimension, that is, its role as an element embedded within the infrastructure of the AI economy. The opaque is not invisible. It reveals only what is strategically convenient to reveal. The narrative of an ethereal AI economy does not deny its extractive character; rather, it stretches and reframes that extraction, relocating it within an apparently ecologically balanced context through images such as “the cloud” and “just code.”

A situation of partial and deliberate lack of transparency, combined with active ignorance, contributes to what can be described as the construction of subcitizenship (Souza, 2003: 178) at the level of global geopolitics. This expanding subcitizenship in the global periphery reflects the form of citizenship produced through exploitative North–South relations.

The inability to exercise informed and context-sensitive decision-making regarding the AI economy, effectively removing part of global society from self-determination in processes of data extraction and technological deployment, crystallizes an exclusionary and precarious subcitizenship in the Global South.

Within this context of regional inequality, peripheral territories are often limited to reproducing imported models rather than developing contextually grounded solutions. Their role becomes confined to contributing to the depletion of natural resources within their own territories. This dynamic unfolds across multiple scales: in the marginalization of Latin America and Brazil (Intercept, 2024), and within Brazil itself, particularly in the Northeastern region (Chamber of Deputies, 2014), where rare earth elements and minerals necessary for the AI economy are extracted without corresponding returns in terms of value-added economic development.

From this perspective, the global AI economy is intimately connected to local and regional extraction. It produces global, unidirectional flows of profit, alongside equally unidirectional flows of local and regional damage.

In this sense, the global ecological crisis, while a genuine challenge, is also appropriated to legitimize the power of extractive corporate conglomerates, further reinforcing the techno-idealism previously mentioned. Before addressing planetary repair, attention shifts toward space exploration. The material impact of technology reveals a disjunction between its capacity to resolve problems and the problems it generates. If technology does not provide a solution, its proponents may argue that the problem itself is not real, which is a fallacy.

3. Transparency in Environmental Impacts: What, How, and For Whom?

Considering the elements of opacity outlined above, effective AI regulation, combined with robust governance measures, can play a fundamental role in promoting transparency and mitigating the negative environmental impacts associated with the AI industry.

These impacts often occur within significant opacity, obscuring both consumption practices and the environmental consequences of AI production and deployment. Opacity in this context is sustained by a lack of transparency in development, implementation, and operational processes, resulting from factors already mentioned that hinder understanding and evaluation of environmental consequences.

The challenges associated with disclosing sensitive information about algorithms and data reflect broader challenges in regulating informational infrastructure. It is necessary to recognize that disclosure may raise legitimate concerns regarding privacy, security, and sovereignty. Therefore, an appropriate balance must be established between transparency and data protection, through policies that safeguard sensitive information while ensuring accountability.

Transparency is a fundamental tool for reducing risks associated with AI. By disclosing information about component origins and environmental impacts, companies are pressured to adopt more sustainable and responsible practices. This not only affects corporate reputation but also strengthens consumer and stakeholder trust. Access to information regarding energy consumption and water use in data centers, for example, can increase public awareness and influence public decision-making concerning these installations.

Supply chains constitute one of the primary areas where transparency is required, as illustrated by the cases of Intel and Apple. Both companies implemented audit mechanisms to certify “conflict-free minerals.” However, these mechanisms have been criticized for circumventing meaningful oversight, relying on audits that are not always conducted independently and that focus on smelters rather than on mining sites themselves, raising concerns about reliability (An Open Letter, 2014; Open Letter: Conflict Minerals, 2014).

Governance mechanisms such as conflict mineral certifications may function as transparency tools, yet opacity emerges when they are manipulated or bypassed to serve corporate interests at the expense of ethical and environmental considerations. Only enough is disclosed to create an appearance of compliance. The weakening of such mechanisms has been described through concepts such as ethical washing, symbolic reform, or, in environmental contexts, greenwashing (Nóbrega and Varon, 2020).

It is essential not only to demand transparency in AI corporate operations, but also to require structural changes in business models and operational practices aimed at mitigating environmental harm.

Transparency alone may not suffice to ensure accountability and justice in AI deployment. Information disclosure must be accessible and comprehensible to all stakeholders, preventing the perpetuation of power asymmetries and informational inequality.

Detailed information about supply chains, including mineral origins and labor conditions in mines and factories, is essential. Clear disclosure of mitigation efforts and environmental policies adopted by companies is equally necessary.

Such measures may include annual environmental reports, independent audits, and real-time publication of resource use and emissions data. Detailed information should be accessible to regulators, investors, researchers, and the public. Transparency must be inclusive, ensuring that communities directly affected by AI operations have meaningful access to information and the ability to participate in decisions concerning technological projects and policies that affect them. In Brazil, many mining sites are located in Indigenous territories without proper consultation.

Transparent and accessible disclosure must extend beyond technical aspects to encompass ethical and social considerations. This includes explaining algorithmic decision-making criteria, identifying and discussing embedded biases, and reflecting on the broader implications of AI applications across different social and cultural contexts.

Abeba Birhane (2021: 6), drawing from relational ethics, emphasizes the importance of moving beyond mere data disclosure when addressing AI impacts. Beyond revealing algorithms and datasets, it is necessary to provide a thorough analysis of potential social, environmental, and ethical consequences. This analysis must consider not only immediate environmental effects but also long-term and systemic consequences arising from interactions between AI systems and social contexts.

When discussing the right to a healthy environment, the focus extends beyond natural elements to the overall quality of lived environments. Birhane’s reflections (2021: 7) underscore the need to consider contextual and disproportionate impacts on marginalized populations when defining what transparency should entail.

Active participation of affected communities and minority groups is essential. Access to AI-related information must be meaningful, inclusive, and responsive to diverse concerns. Transparency should not be treated as an end in itself, but as part of a normative framework that establishes specific obligations concerning accountability, justice, and social well-being.

Finally, the business model dimension itself, once made visible through transparency, warrants scrutiny and collective deliberation. If a business model inherently contributes to planetary environmental degradation, can its operation proceed without the strictest precautionary measures?

Conclusions

The analysis presented here underscores the urgency of critically addressing the socio-environmental impacts and regulatory dynamics of artificial intelligence. The opacity permeating AI supply chains and production processes perpetuates extractive and unsustainable practices while concealing their true environmental and human costs.

Opacity within the AI industry is multidimensional. Natural resource extraction, massive data center energy consumption, and precarious labor are frequently concealed beneath narratives of clean digital innovation. This narrative is not accidental; it is a deliberate strategy to obscure real costs and protect corporate interests. The establishment of transparency mechanisms is therefore essential to expand the social accountability of these businesses and enable proper assessment of AI impacts.

Robust regulation, independent audits, and detailed disclosure of supply chains and resource consumption are concrete measures to promote transparency. Corporate accountability must include structural changes in business models aimed at mitigating harm and fostering sustainable practices. Transparency must be inclusive and meaningful, ensuring access to information for all affected stakeholders.

Regional specificity cannot be ignored. In Brazil, rare mineral extraction frequently occurs in Indigenous and vulnerable territories, intensifying inequality and injustice. This structural subcitizenship, in which significant portions of society are excluded from decision-making and economic benefits, reflects broader global dynamics of power and exploitation. As Malcolm Ferdinand (2022) argues, this constitutes a double fracture linking environmental destruction and colonial legacy.

Ultimately, efforts to ensure transparency and accountability in the AI industry must be guided by principles of equity and sustainability. Environmental and social harms must not disproportionately burden vulnerable communities. Only through a genuine commitment to climate justice can technological development align with environmental limits and human dignity.

REFERENCES

AN OPPER LETTER. Publicado em 09/09/2014. Acesso em 25/04/2024. Disponível em: <https://suluhu.org/wp-content/uploads/2014/10/09092014- open-letter-final-and-list-doc.pdf>

BIRHANE, Abeba. Algorithmic injustice: a relational ethics approach. Patterns, v. 2, n. 2, publicado em 12/02/2021. Acesso em 14/05/2024. Disponível em: <https://doi.org/10.1016/j.patter.2021.100205>.

BRASIL. Supremo Tribunal Federal. Ação Direta de Inconstitucionalidade nº 4277, Requerente Procurador Geral da República, Ministro Relator Ayres Britto, 5 de maio de 2011, publicado em 14/10/2011, Disponível em: <http:// redir.stf.jus.br/estfvisualizadorpub/jsp/consultarprocessoeletronico/ ConsultarProcessoEletronico.jsf?seqobjetoincidente=11872>. Acesso em 22 abril 2020.

BURKE, Peter. Ignorância: Uma História Global. São Paulo: Vestígio, 2024.

CÂMARA DOS DEPUTADOS. Minerais estratégicos e terras-raras [recurso eletrônico]. Relator Colbert Martins; Paulo César Ribeiro Lima … [et al.]. – Brasília: Câmara dos Deputados, Edições Câmara, 2014.

CAMARGOS, Daniel. Exclusivo: Apple, Google, Microsoft e Amazon usaram ouro ilegal de terras indígenas brasileiras. Repórter Brasil (online). Publicado em 25/07/2022. Acesso em 24/04/2024. Disponível em: https://reporterbrasil. org.br/2022/07/exclusivo-apple-google-microsoft-e-amazon-usaram-ouroilegal-de-terras-indigenas-brasileiras/

CRAWFORD, Kate. Atlas of AI: Power, Politics and The Planetary Costs of Artificial Intelligence. New Haven and London: Yale University Press, 2021.

DIAS, Tatiana; MOTORYN, Paulo. Lítio, X e Eleições: Seguimos o dinheiro que movimenta os ataques de Elon Musk a Alexandre de Moraes. INTERCEPT (online). Publicado em 08/04/2024. Acesso em 28/05/2024. Disponível em:https://www.intercept.com.br/2024/04/08/seguimos-o-dinheiro-quemovimenta-os-ataques-de-elon-musk-a-alexandre-de-moraes/

FERDINAND, Malcom. Uma Ecologia Decolonial: Pensar a Partir do Mundo Caribenho. São Paulo: Ubu Editora, 2022.

FERRARI, Vanessa. Inteligência Artificial: Um Paradoxo Ambiental? Cadernos Jurídicos, São Paulo, ano 24, nº 65, p. 201-212, Janeiro-Março, 2023.

FRITSCH, K., & VAN DER WAAL, S. Meeting at the intersection: Sharing resources to expose tech solutionism in climate narratives. Superrr Digital Futures. 2022.

MICHAELS, David. O Triunfo da Dúvida: dinheiro obscuro e a ciência da enganação. São Paulo: Elefante / O Joio e o Trigo, 2024.

MONTANINI, Marcelo. Um país essencial para alta tecnologia, mas afundado em guerra. Nexo Jornal (online). Publicado em 06/05/2024. Acesso em 23/05/2025. Disponível em: https://www.nexojornal.com.br/expresso/2024/05/06/umpais-essencial-para-alta-tecnologia-mas-afundado-em-guerra.

NÓBREGA, Camila; VARON, Joana. Big tech goes green(washing): Feminist lenses to unveil new tools in the master’s houses. In Global Information Society Watch 2020 – Technology, the Environment and a sustainable world: Answers from the Global South. Association for Progressive Communications – APC, 2021. Disponível em https://giswatch.org/node/6254#_ftn12

O’BRIEN, Isabel. Data center emissions probably 662% higher than big tech claims. Can it keep up the ruse? The Guardian (online). Publicado em 15/12/2024. Acesso em 23/05/2025. Disponível em: https://www.theguardian. com/technology/2024/sep/15/data-center-gas-emissions-tech

OPEN LETTER: CONFLICT MINERALS: A BROADER PUSH FOR REFORM IS ESSENTIAL. Publicado em 30/10/2014. Acesso em 03/06/2024. Disponível em: <https://enoughproject.org/files/OpenLetterConflictMinerals_October_2014. pdf>.

PEÑA, Paz. ¿Qué diablos es la IA sostenible? Instituto Latinoamericano de Terraformación. Santiago de Chile, 2025.

PROCTOR, Robert N. Agnotology: A Missing Term to Describe the Cultural Production of Ignorance (And Its Study). In
PROCTOR; SCHIEBINGER (org). Agnotology: The Making and Unmaking of Ignorance. California: Stanford University Press, 2008.

RADWIN, Maxwell. Shady contracts, backdoor deals spur illegal gold mining in Bolivian Amazon. Mongabay (online). Publicado em 08/11/2022. Acesso em 25/06/2025. Disponível em: https://news.mongabay.com/2022/11/shadycontracts-backdoor-deals-spur-illegal-gold-mining-in-bolivian-amazon/.

RATHI, Akshat; WHITE, Natasha. How Tech Companies Are Obscuring AI’s Real Carbon Footprint. Bloomberg (online). Publicado em 21/08/2024. Acesso em 23/05/2025. Disponível em: https://www.bloomberg.com/news/ articles/2024-08-21/ai-tech-giants-hide-dirty-energy-with-outdatedcarbon-accounting-rules

ROLLEY, Sonia. Congo files criminal complaints against Apple in Europe over conflict minerals. Reuters (online). Publicado em 17/12/2024. Acesso em 25/05/2025. Disponível em https://www.reuters.com/sustainability/societyequity/congo-files-criminal-complaints-against-apple-europe-over-conflictminerals-2024-12-17/.

SILVA, Tarcízio. Racismo Algoritmico: Inteligência Artificial e Discriminação nas Redes Digitais. São Paulo: Edições Sesc São Paulo, 2022.

SOUZA, Jessé. A construção social da subcidadania: para uma Sociologia Política da modernidade periférica. Belo Horizonte: UFMG, 2003.

TEIXEIRA, Pedro S. Data center pode consumir energia de cidade de 30 mil habitantes, e IA vai aumentar gasto. Folha de São Paulo (online). Publicado em 30/03/2024. Acesso em 24/04/2024. Disponível em: https://www1.folha.uol. com.br/tec/2024/03/data-center-pode-consumir-energia-de-cidade-de-30- mil-habitantes-e-ia-vai-aumentar-gasto.shtml

André Fernandes