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, Anicely Santos, Helton Leyendecker, Carolina Branco, Clarissa Mendes, Ramon D´barssoles, Rhaiana Valois

Problem description

With the emergence of cloud computing technologies, the massive production of data and, especially, the accelerated adoption of Artificial Intelligence (AI) in society, the demand for data centers has become critical, resulting in the urgent need to strengthen the storage and processing capacities of these infrastructures. A data center is a physical space where information is processed, stored, transmitted, and managed, constituting the core of modern Information Technology (IT) infrastructure. The Synergy Research Group indicates that the number of large-scale data centers in 2024 surpassed 1,000, representing 41% of the world’s data processing capacity. The expectation is that by 2029 this percentage will reach 60% [1]. For all this infrastructure to achieve the operational capacity required to meet these demands, the following factors must be considered: (i) dependence on brown energy [2], generated from polluting sources such as coal and oil; (ii) massive CO₂ emissions; (iii) intensive mineral extraction for the manufacture of physical components; (iv) the generation of electronic waste and other residues; and (v) high water consumption.
The energy consumption of a data center is distributed across different systems, with approximately 50% allocated to IT infrastructure, which includes servers, storage, and networks. Air conditioning and cooling systems account for 37% of the total. In addition, 10% of consumption is attributed to power distribution systems, while auxiliary lighting systems correspond to 3%.
In the global context, over the last decade, data centers have been responsible for 2.4% of the total electricity consumption worldwide [3]. In the short term, according to the consulting firm Oliver Wyman, global demand is expected to grow by 16% by 2026. To illustrate, a single data center company in São Paulo can consume energy equivalent to that of a municipality with 150,000 inhabitants [4]. Regarding CO₂ emissions, these infrastructures currently account for 0.3% of global carbon emissions [5], in addition to indirect emissions. For example, the manufacture of each server produces approximately 1,300 kg of CO₂ [6]. Thus, the growing production of servers to supply these data centers is capable of releasing millions of tons of CO₂ into the planet’s atmosphere. Although companies invest in carbon offset markets as a strategy to mitigate global warming, they continue to release greenhouse gases into the atmosphere. This creates the impression that it is possible to keep stimulating the accelerated growth of this technology through an extractivist and predatory logic that, in practice, does not reduce emissions at their source.
Projects in the carbon offset market tend to target areas with low deforestation rates and higher regenerative capacity, such as Indigenous lands. In this way, the implementation of these programs can generate conflicts and forced displacement, since decisions about land use often ignore the interests and ways of life of these populations. To illustrate the scale, the largest data center in Latin America, located in Vinhedo (São Paulo), occupies 46,000 m² of land [7], equivalent to nearly seven football fields. The largest in the world, located in China, covers one million square meters [8]—that is, about 140 football fields. Furthermore, the manufacturing of the physical components of these infrastructures requires large quantities of mineral resources. Beyond direct environmental impacts, this may also generate health problems for local communities [9]. Additionally, the maintenance of these data centers generates a considerable amount of electronic waste, as well as other residues that, when improperly disposed of, increase environmental pollution. Despite initiatives aimed at reusing these components, a large portion of these materials is still not properly recycled [10]. Finally, regarding water consumption, water is used to cool servers in order to keep them operating at appropriate temperatures. To illustrate, generating a simple 100-word text in OpenAI’s GPT-4 requires servers to use the equivalent of three bottles of water for cooling [11]. The intensive use of this finite resource may worsen water scarcity in regions where water demand already exceeds natural supply, as occurs in areas experiencing high water stress [12]. In Brazil, where more than half of the rivers are drying up [13], the installation of these data centers may place additional pressure on already strained water resources.

2. Measuring Consumption and Efficiency in Data Centers

IT equipment and cooling systems are the two main components, responsible for approximately 90% of the total energy consumption of data centers. Server operation generates heat, resulting in a cooling load on refrigeration equipment, which increases energy consumption. Analyzing how much energy each server consumes during peak periods helps balance the high processing demand across servers. There are some metrics to support these analyses, which have been evaluated by researchers focusing on the optimization of IT equipment and cooling systems [14]. Among these, Power Usage Effectiveness (PUE) stands out. PUE indicates the ratio between the total energy required to operate a data center (including components such as cooling) and the energy used solely by computing devices. A PUE of 1.0 implies ideal efficiency, meaning that the data center uses only the energy required to power the computing devices. To better understand, a PUE of 1.2 indicates that for every 1 kWh used by computing hardware, the total energy consumption of the data center is 1.2 kWh. Globally, the average PUE for data centers in 2023 was 1.58.
Before the pandemic, large-scale data centers consumed between 5 and 10 megawatts (MW) of energy. Since 2022, large facilities have reached between 20 MW and 50 MW, a growth of about 300%, highlighting a demand for energy resources without a corresponding increase in energy supply. In Brazil, the total of these facilities has consumed approximately 580 MW, with expectations of exceeding 2 gigawatts (GW) by 2028. This figure excludes the energy used for cooling, which represents about 50% of the total energy consumption of a data center [15].

2.1 The scenario in some countries

In 2016, China established the GB/T32910-2016 standard, which redefines Electric Energy Usage Effectiveness (EEUE) in relation to PUE [16], significantly compensating for deficiencies in domestic electrical efficiency standards for data centers. EEUE is identical to PUE in its calculation formula, considering 1 as the maximum level of energy efficiency; however, this is understood as an ideal condition that cannot realistically be achieved. At the same time, when calculating EEUE, considering differences caused by cooling technologies, load utilization rates, data center classification, and the regional climate environment, a correction metric called EEUE-X is proposed to compensate for systemic differences. Under this metric, data centers are divided into five categories according to their energy efficiency values: energy-saving (EEUE up to 1.6), relatively energy-saving (EEUE up to 1.8), qualified (EEUE up to 2.0), relatively energy-consuming (EEUE up to 2.2), and highly energy-consuming (greater than 2.2). Additionally, the standard also considers that data centers operating at security levels crucial to national security and social order include factors such as compressors, humidification, fresh air systems, power supply, and lighting, among others, which impact the EEUE-X value. This implies that high-security data centers will not be disproportionately penalized in energy efficiency assessments due to their higher consumption required to keep these critical systems operating continuously and redundantly.
In Singapore [17], due to its tropical climate, the country implemented the world’s first “tropical” standards for data centers. Given its geographical location and climate, where temperatures and humidity are generally high, cooling needs represent a significant share of total energy consumption. Singapore’s tropical standards recognize that it is possible to operate at higher temperatures (26°C or above) without compromising equipment reliability. By establishing an optimized operational temperature range for tropical climates, these standards allow for a significant reduction in energy consumption associated with cooling. For each 1°C increase in operational temperature, there is a potential energy saving ranging from 2% to 5%. This demonstrates the substantial impact that optimizing operational temperature can have on the overall energy efficiency of data centers in tropical climates. In 2023, Germany established an Energy Efficiency Act that implements the Energy Efficiency Directive (EED) [18], representing an improvement over what is established by the European Union (EU). This law extends reporting obligations to smaller data centers, with installed capacity starting at 300 kW, while the EU directive applies to data centers with power demand of at least 500 kW. The law also expands the obligation to implement an energy management system for data centers and clients with equipment demanding more than 50 kW. Regarding PUE, it must range between 1.5 and 1.2 depending on the age of the data center, and the Energy Reuse Factor (ERF) must range between 10% and 20%. Additionally, data centers are required to operate with 50% renewable energy, increasing this requirement to 100% by January 1, 2027. Another requirement is that data center operators must annually inform their clients about the energy consumption directly attributable to them. This measure is seen as an important prerequisite for companies developing AI to optimize their energy consumption.

3. National context

Brazil has become a strategic hub for the installation of data centers, attracting major international investments due to its geological stability and renewable energy potential [19]. These factors, combined with government incentives [20], place the country in a prominent position in Latin America. However, fostering this position may also bring the risk of deep socio-environmental impacts that have not been adequately addressed, both in terms of national regulation and in terms of dialogue with society. Although Brazil has a largely renewable energy matrix, considering its hydroelectric, solar, and wind generation capacity, most of these sources are intermittent in nature—that is, energy availability is neither stable nor constant. In contrast, the operation of data centers requires continuity and operational stability, meaning that ensuring this energy consistency implies the need to use non-renewable energy sources [21]. Furthermore, at the national level, there is no sufficient regulatory framework to address and prevent practices such as greenwashing, which consists, among other actions, of increasing investment in renewable energy in order to justify increased use, elsewhere in the operational chain, of non-renewable energy. Beyond the energy issue, predatory mining for hardware production is a pressing issue that lacks sufficient legislative attention. The manufacture of servers requires massive quantities of lithium [22], copper [23], and cobalt [24], among other strategic minerals essential for the production of high-technology products (known as rare earths [25]), including gold, which is extracted in Brazil under environmentally and socially devastating conditions [26]. Illegal mining, especially in the Amazon, already causes deforestation, river contamination, and violence against Indigenous peoples—as seen in Yanomami territories, where illegal mining is associated with malaria outbreaks and malnutrition [27]. While the growing demand for data centers already represents a significant sustainability challenge, the expansion of AI exponentially worsens this scenario. While data centers traditionally operating cloud services use between 20 and 30 MW, those dedicated to AI require between 150 and 200 MW, with some reaching the gigawatt scale [28].
Despite this significant impact, Bill (PL) 2338/23, the main legislative proposal currently under discussion to regulate AI in the country, addresses environmental issues only superficially, dedicating just one provision to sustainability topics. This legislative gap reflects the lack of an integrated approach to mitigating the environmental effects of technological expansion.

4. Guidelines – Bill 2338: a possible but insufficient starting point

There is still no specific regulation for data centers in Brazil, nor are there standards that address environmental implications in detail. Issues such as energy efficiency, the use of renewable sources, and impact mitigation remain without clear guidelines. In the current scenario, the only regulatory reference that touches on this topic is Bill 2338/23, which, although focused on AI regulation, makes a passing reference to “sustainable data centers.” This proposal represents a starting point for the development of a more comprehensive regulatory framework. The mentioned provision in the bill is Article 59, item IV, as follows:

Art. 59. Public administration at the federal, state, Federal District, and municipal levels may promote innovation and productive and technological development in AI.
Sole paragraph. The support measures referred to in the caput shall be guided by the following directives:

[…]
IV – encouraging the expansion of the availability of sustainable data centers with high data processing capacity for AI systems, with the densification of this production chain and related digital services in Brazil, with the objective of supporting the productive sector and technical-scientific research and development;
Despite the aforementioned reference to the need to foster “sustainable data centers,” this guideline still lacks detail and practical effectiveness. For AI regulation in Brazil to be aligned with a responsible development model, it is essential to adopt concrete measures that ensure the mitigation of the environmental impacts of this sector.
In view of this, and considering the elements highlighted throughout this policy brief, the following additions to Article 59 of Bill 2338/23 are proposed, aiming to ensure greater transparency and effectiveness in the implementation of policies aimed at the sustainability of data centers:
Inclusion, in item IV, of subparagraph (a), as follows:
a) For the purposes of item IV, sustainable data centers are considered to be those that adopt concrete measures to mitigate environmental impacts, including, but not limited to:
i) priority use of energy from renewable sources;
ii) implementation of cooling systems with high energy efficiency and lower water consumption;
iii) compliance with standardized metrics for measuring and mitigating carbon footprint;
iv) effective policies for the reuse and sustainable disposal of hardware.
Inclusion of the following items:
V – establishing standard metrics for environmental resource consumption for data centers and servers, regardless of size;
VI – having energy measurement equipment, with measurement carried out at the server level rather than only at the data center level;
VII – conducting environmental risk and impact assessments in the locations where data centers are or will be deployed, making public reports available on the consumption of environmental resources;
VIII – regulating AI training plans in national data centers, including the possibility of limiting AI training to periods of lower energy demand;
IX – data center and server companies must implement renewable energy sources to offset excessive energy use;
X – adopting independent verification and certification mechanisms to prevent “greenwashing” practices, ensuring that environmental claims are auditable and based on objective and measurable criteria.

5. Conclusion

The accelerated expansion of AI and the resulting increase in demand for data centers pose significant environmental challenges. High energy consumption, CO₂ emissions, intensive water use, electronic waste, and the impacts of resource extraction for hardware production are issues that require urgent regulatory attention. Although Brazil has significant renewable energy potential, the absence of regulation compromises the possibility of responsible growth as a data center hub. Bill 2338/23 represents an initial step by mentioning the need to foster sustainable data centers, but it is still insufficient to ensure concrete measures to mitigate environmental impacts. The implementation of standardized metrics to monitor resource consumption, the requirement of environmental impact assessments, and the adoption of renewable energy sources combined with mechanisms to mitigate greenwashing practices are essential measures to ensure sustainable technological development. These proposals are aligned with the international commitments assumed by Brazil, particularly the United Nations Sustainable Development Goals. A stronger regulatory framework that effectively establishes sustainability as a primary guideline in technological development goes beyond an opportunity for leadership in the digital economy: it is an indispensable condition given the planet’s biophysical limits. Due to its cumulative impact on the climate crisis and the pressure it places on finite natural resources, the sector requires rules that translate this urgency into concrete mechanisms for oversight and damage mitigation. The alternative—unregulated technological development—has the potential to generate conflicts in the distribution of water and energy resources, increased CO₂ emissions, and the deepening of socio-environmental problems related to predatory mining, among other risks. Thus, it is essential that the regulatory debate on sustainability and data centers be expanded and that national legislation incorporate stricter guidelines to reconcile digital growth with environmental preservation. Only with a robust and transparent regulatory framework will Brazil be able to consolidate itself as a leader in the adoption of sustainable practices in the digital infrastructure sector.

References
[1] Synergy Research Group. Hyperscale Operators and Colocation Continue to Drive Huge Changes in Data Center Capacity Trends. 2024. Disponível em: https://www.srgresearch.com/articles/hyperscale-operators-and-colocation-continue-to-drive-huge-changes-in-data-center-capacity-trends. Acesso em: 25/03/2025.
[2] T. Bhattacharya and X. Qin, “Modeling Energy Efficiency of Future Green Data centers,” 2020 11th International Green and Sustainable Computing Workshops (IGSC), Pullman, WA, USA, 2020, pp. 1-3.
[3] Brito, José Luiz & Matai, Patricia & Santos, Mario. (2023). Data Center e Eficiência Energética. Brazilian Journal of Business. 5. 786-795. 10.34140/bjbv5n2-002.
[4] Portal G1. Brasil é líder em investimentos em data centers na América Latina; cidade na Grande SP tem quinto maior complexo do planeta. 2023. Disponível em: https://g1.globo.com/jornal-da-globo/ noticia/2023/09/23/brasil-e-lider-em-investimentos-em-data-centers-na-america-latina-cidade-na- -grande-sp-tem-quinto-maior-complexo-do-planeta.ghtml. Acesso em: 25/03/2025.
[5] Estampa. Cartography of generative AI. 2024. Disponível em: https://cartography-of-generative-ai. net/. Acesso em: 25/03/2025.
[6] SAR, Yusuf. The Silent Burden Of AI: Unveiling The Hidden Environmental Costs Of Data Centers By 2030. Forbes. 2024. Disponível em: https://www.forbes.com/councils/forbestechcouncil/2024/08/16/the-silent-burden-of-ai-unveiling-the-hidden-environmental-costs-of-data-centers- -by-2030/. Acesso em: 25/03/2025
[7] Ascenty. Qual o maior Data Center da América Latina? 2024. Disponível em: https://ascenty.com/ blog/artigos/maior-data-center-da-america-latina/#:~:text=O%20Data%20Center%20de%20Vinhedo,universidades%20e%20institui%C3%A7%C3%B5es%20educacionais. Acesso em: 25/03/2025.
[8] Greenergy Data Centers. Where is the world’s largest data center? 2022. Disponível em: https:// www.greenergydatacenters.com/eng/blog/where-is-the-worlds-largest-data-center. Acesso em: 25/03/2025.
[9] Agência Gov. Yanomamis de nove aldeias estão contaminados por mercúrio. 2024. Disponível em: https://agenciagov.ebc.com.br/noticias/202404/yanomamis-de-nove-aldeias-assediadas-pelo-garimpo-estao-contaminados-por-mercurio. Acesso em: 25/03/2025.
[10] OECD (2022), “Measuring the environmental impacts of artificial intelligence compute and applications: The AI footprint”, OECD Digital Economy Papers, No. 341, OECD Publishing, Paris, https://doi. org/10.1787/7babf571-en.
[11] GONÇALVES, André Luiz Dias. GPT-4 gasta até 3 garrafas de água a cada 100 palavras geradas, diz estudo. Tecmundo. 2024. Disponível em: https://www.tecmundo.com.br/mercado/289830-gpt-4-gasta-3-garrafas-agua-cada-100-palavras-geradas-diz-estudo.htm?ab=true&. Acesso em: 25/03/2025.
[12] TAQI, Joanne Emerson; JOHNSON, James. Data centers and water: From scrutiny to opportunity. White Case. 2024. Disponível em: https://www.whitecase.com/insight-our-thinking/data-centers-and-water-scrutiny-opportunity#:~:text=A%20small%20one%2Dmegawatt%20data,stressed%20watersheds%20where%20the%20demand. Acesso em: 25/03/2025.
[13] BATAIER, Carolina. Mais da metade dos rios brasileiros está secando; problema é maior onde há atividade agrícola intensiva. Brasil de Fato. 2025. Disponível em: https://www.brasildefato.com. br/2025/02/17/mais-da-metade-dos-rios-brasileiros-esta-secando-problema-e-maior-onde-ha-atividade-agricola-intensiva/. Acesso em: 25/05/2025.
[14] SHAO, Xiaotong; ZHANG, Zhongbin; SONG, Ping; FENG, Yanzhen; WANG, Xiaolin. A review of energy efficiency evaluation metrics for data centers. Energy and Buildings, v. 271, p. 112308, 2022. https://doi.org/10.1016/j.enbuild.2022.112308.
[15] VARGAS, Ivan Martínez. Brasil já colhe benefícios do boom dos data centers. Valor. 2024. Disponível em: https://valor.globo.com/publicacoes/especiais/investimento-estrangeiro/noticia/2024/11/06/ brasil-ja-colhe-beneficios-do-boom-dos-data-centers.ghtml. Acesso em: 25/03/2025.
[16] SHAO, Xiaotong; ZHANG, Zhongbin; SONG, Ping; FENG, Yanzhen; WANG, Xiaolin. Op. Cit.
[17] SINGAPORE. MINISTRY OF COMMUNICATIONS AND INFORMATION; INFO-COMMUNICATIONS MEDIA DEVELOPMENT AUTHORITY. Singapore’s Digital Connectivity Blueprint. Singapura, 5 jun. 2023. Acesso em: 27/03/2025.
[18] EBERT, Kai; ALDER, Nicolas; HERBRICH, Ralf; HACKER, Philipp. AI, Climate, and Regulation: From Data Centers to the AI Act. 2024. Preprint arXiv:2410.06681. Disponível em: https://arxiv.org/ abs/2410.06681. Acesso em: 25 mar. 2025.
[19] STORCH, Julia. Por que o Brasil é o principal destino Latam de data centers? Exame. 2022. Disponível em: https://classic.exame.com/bussola/por-que-o-brasil-e-o-principal-destino-latam-de-data- -centers/?utm_source=copiaecola&utm_medium=compartilhamento. Acesso em: 25/03/2025.
[20] Portal do Governo Federal. PBIA prevê R$ 500 milhões para data centers verdes, que aliam tecnologia e sustentabilidade. 2025. Disponível em: https://www.gov.br/mcti/pt-br/acompanhe-o-mcti/ noticias/2025/02/pbia-preve-r-500-milhoes-para-data-centers-verdes-que-aliam-tecnologia-e-sustentabilidade. Acesso em: 25/05/2025.
[21] Portal da Câmara. Soberania nacional e a política de data centers no Brasil. 2024. Disponível em: https://www.camara.leg.br/radio/programas/1116955-ep91-soberania-nacional-e-a-politica-de- -data-centers-no-brasil/. Acesso em: 25/03/2025.
[22] NEVES, Ernesto. Brasil se lança na corrida pela produção do lítio, o mineral do futuro. Veja. 2023. Disponível em: https://veja.abril.com.br/agenda-verde/brasil-se-lanca-na-corrida-pela-producao-do-litio-o-mineral-do-futuro/. Acesso em: 25/03/2025.
[23] VENDITTI, Bruno. Why Copper Is Critical for Data Centers. Elements. 2023. Disponível em: https://elements.visualcapitalist.com/why-copper-is-critical-for-data-centers/. Acesso em: 25/03/2025.
[24] ZHENG, March. The Environmental Impacts of Lithium and Cobalt Mining. Earth.org. 2023. Disponível em: https://earth.org/lithium-and-cobalt-mining/. Acesso em: 25/03/2025.
[25] Câmara dos Deputados. Minerais Estratégicos e Terras-Raras. Centro de Estudos e Debates Estratégicos. Série Estudos Estratégicos n. 3. Edições Câmara. Brasília, 2014. Disponível em: https:// www2.camara.leg.br/a-camara/estruturaadm/altosestudos/pdf/minerais-estrategicos-e-terras-raras. Acesso em: 25/03/2025.
[26] Centro de Informação sobre Empresas e Direitos Humanos. Brasil: Multinacionais podem estar usando ouro ligado ao garimpo ilegal e a violações de direitos indígenas em suas cadeias de produção; incl. comentários das empresas. 2024. Disponível em: https://www.business-humanrights.org/ pt/%C3%BAltimas-not%C3%ADcias/brasil-multinacionais-podem-estar-usando-ouro-ligado-ao-garimpo-ilegal-e-a-viola%C3%A7%C3%B5es-de-direitos-ind%C3%ADgenas-em-suas-cadeias-de-produ%C3%A7%C3%A3o-incl-coment%C3%A1rios-das-empresas/. Acesso em: 25/03/2025.
[27] JUNGMANN, Luiza Greenhalgh. Ecocídio e genocídio privados na Terra Indígena Yanomami. Consultor Jurídico. Disponível em: https://www.conjur.com.br/2024-nov-03/ecocidio-e-genocidio-privados-na-terra-indigena-yanomami/. Acesso em: 25/03/2025. [28] RYNGELBLUM, Ivan. Brasil avança em data centers e caminha para se tornar a nova fronteira do setor. Neofeed. 2025. Disponível em: https://neofeed.com.br/negocios/brasil-avanca-em-data-centers-e-caminha-para-se-tornar-a-nova-fronteira-do-setor/. Acesso em: 25/03/2025.

Institucional