The Environmental Externality of Artificial Intelligence: Regulatory Gaps in AI’s Environmental Oversight and Proposed Legal Reforms

I. Introduction

Artificial intelligence has rapidly become embedded in nearly every sector of the American economy. From consumer-facing chatbots to national defense applications, AI systems promise unprecedented efficiency and economic growth. On January 23, 2025, President Trump issued a sweeping executive order, Removing Barriers to American Leadership in Artificial Intelligence. In Section 1. Purpose. of this order, President Trump, “revokes certain existing AI policies and directives that act as barriers to American AI innovation, clearing a path for the United States to act decisively to retain global leadership in artificial intelligence.” The federal government has actively encouraged this expansion in order to remain at the forefront of innovation and to stay competitive globally, but at what cost? Beneath this novelty tool lies a largely overlooked environmental toll. Data centers require immense water consumption, electricity demand, and carbon emissions in order to train and use these large-scale AI models. Existing environmental statutes, not designed to regulate AI infrastructure at scale, produce an inadequate legal framework that demands targeted reform. Thus, AI developers should be treated as major industrial environmental actors under federal environmental law and subject to mandatory carbon and water-use disclosure requirements, and the government must propose legal reforms to mediate environmental damage.

II. The Environmental Footprint of AI

A. Water Consumption

AI developers like Open AI use data centers to train and program AI bots such as ChatGPT. These processes use significant amounts of energy which are then converted into heat; the heated machinery then requires water for cooldown. The AI model also uses a significant amount of water in its reasoning process—when a user asks it to answer a question or to generate text. A simple conversation of 20-50 questions uses about the same amount of water in the average 16oz. water bottle. Now that AI is open for public use, the scale of water usage is larger than ever, with OpenAI admitting that ChatGPT currently receives about 2.5 billion prompts from users daily.

B. Energy Use & Carbon Emissions

Google Gemini is the first result on most Google searches these days, and many people have transitioned from using regular search engines, to directly looking up information on ChatGPT or other AI models. Researchers run the risk of using unidentified or proofed sources through this method, but the planet faces an even direr consequence. Each prompt on ChatGPT consumes about ten times as much energy used for a Google search. Estimates state that ChatGPT’s daily consumption of energy is enough to charge eight million smartphones. Generative AI is now the largest category of computing facilities worldwide, with 54% of the global capacity in the United States. Those data centers in the US that make up more than half of the world’s capacity are using approximately the same amount of energy that all of the residential housing requires nationwide. The International Energy Agency (IEA) predicts the global electricity demand from data centers will at least double in the next five years. 

While renewable energy is popularizing, the U.S. grid still relies on fossil fuels, so as electricity use by data centers increases, so will greenhouse gas emissions at an exponential rate. Even without accounting for emissions from everyday global usage, the training process for these models alone requires a considerable amount of electricity. For example, training GPT-3 consumed around 1,287 megawatt-hours of electricity and produced 552 tons of carbon dioxide. Outside of training, it is estimated that ChatGPT generates 8.4 tons of carbon dioxide every year.

C.Ethical and Societal Considerations

In addition to consuming valuable resources and harmful carbon emissions, the environmental impact of AI presents considerations about the ethics of sustaining the projected growth for AI. The placement of data centers affects marginalized communities and puts a strain on local resources. In Des Moines, Iowa, the Microsoft data centers contribute to 6% of freshwater consumption in the area in the month. According to Deloitte India’s report, Attracting AI Data Centre Infrastructure Investment in India, the country will require an additional 40-50 Terawatt-hours of electricity and 45-50 million square feet of real estate to meet the expected demand for AI data centres by 2030. India is a good location choice as it has emerged as a power surplus nation, yet still, fuel sources are finite and must be used sensibly. Along with looking at the available power supply, it is important to consider the ethics of placing many facilities in each location. When choosing where to plant facilities, the local population’s welfare must be taken into consideration. Will they be at risk for inhaling excessive greenhouse gases? Will the facility impose too much on their water supply? 

III. The Existing Regulatory Framework 

There are existing statutes for industrial facilities to make environmental impact reports to the Environmental Protection Agency (EPA), and for there to be oversight in the environmental risks associated with various kinds of pollution. The Senate recently passed the Artificial Intelligence Environmental Impacts Act of 2024, which contained three main provisions:

1. the administrator of the EPA must conduct an inspection of the environmental impacts of AI

2. the Director of the National Institute of Standards and Technology must assemble an association on said environmental impacts

3. the Director of the NIST must develop a voluntary reporting system for reporting environmental impacts

The Clean Air Act (CAA) is a federal law regulating air emissions from various sources and authorizing the EPA to establish National Ambient Air Quality Standards (NAAQS) to protect public health and manage hazardous emissions. Section 112 of the act addresses the emission of air pollutants which requires technology-based standards to be issued for “major sources” and certain “area sources.” “Major sources” are defined as stationary sources that emit at least 100 tons of any air pollutant per year. "Area sources" are any stationary sources that are not considered a major source. This section of the bill requires that the EPA authorize standards of the maximum degree of reduction in emissions, referred to as the maximum achievable control technology (MACT) standards. Once the MACT standards are issued, the EPA is required to review them after eight years to determine if there are any remnant risks for that source category, and to revise the standards if necessary. The act also aimed to set NAAQS in every state, and having each state develop implementation plans in order to achieve the standards. Title V of the CAA requires any major source or area source to obtain a Title V Permit. These are legally-enforceable documents outlined to improve compliance and clarify which facilities must control air pollution. These permits can be issued by the EPA, but most are issued by state or local agencies.

IV. The Regulatory Gap

There are laws against wasting water, but these are primarily enforced at smaller levels through local and state regulations. These laws include penalties for failing to repair leaks, prohibitions on excessive watering, and restrictions on industrial waste, but typically do not reach the national, institutional level. María Montero from IE University emphasizes that, “in a world where 2.2 billion people still live without access to safe drinking water, this invisible cost is not just unsustainable – it is profoundly unethical.”

The burden of technological innovation must be considered by big tech firms who are not directly affected. A narrow public awareness of operations is permitting the limited accountability of these firms. The individuals in power who are able to make the decisions to be transparent about the scale of the effects of AI are able to be dismissive, because they are not being pressured by the public, or the government. In fact, the public, the government, and the private sector are increasingly integrating AI into their regular procedure. The Artificial Intelligence Environmental Impacts Act of 2024 bill was a huge advancement, yet still, the reporting is not mandated. The Clean Air Act has a greater capacity to mandate a reporting system or even set limitations to control pollution, but currently disclosure policies are incomplete. 

While AI companies are subject to generally applicable environmental statutes, the senate admits that “there are no legally binding energy standards that apply explicitly to operation of data centers in the private sector.” The Clean Air Act states that the EPA may monitor and establish standards for “major sources”. With annual carbon emissions of 8.4 tons, ChatGPT is far below being considered a “major source”, emitting at least 100 tons per year, and therefore left unregulated. However, while the annual carbon emissions are less than the minimum threshold for a “major source”, the training process to develop the chat bots emits more than five times the minimum emission requirement to be considered a “major source” at a whopping 552 tons of carbon dioxide. That was just to train GPT-3, but AI is still developing. More models will be released, and each of them requires training. As of August 2025, OpenAI released its most current model, GPT-5, but there are no figures to disclose how much carbon emissions its training process expelled, as OpenAI does not disclose this data.

V. Proposed Solutions

AI is public, rampant, and unregulated. It has gone viral worldwide and has become woven into the daily lives of individuals, companies, and institutions. It would not be feasible to deintegrate it from public life, but governments can regulate and contain the damage.

Water usage limits could be an option in the future, but it would be difficult to enforce widespread limitations now that AI is becoming integrated into every institution and the government has announced plans to charge towards enhancing international AI dominance. However, mandatory water-use disclosure for AI training should be required. It is vital that AI companies provide transparent reports about their greenhouse gas emission as well. 

Reforming the AI business to maximize sustainability should be a priority and a key step in that process is keeping the public informed. A precedent for this would be the LEED rating system. This certification is the most widely used system for verifying that a building is built to optimize sustainability designs focusing on energy savings, water efficiency, reduced carbon emissions, and improved indoor air quality. LEED was successful because it was able to establish a market competitiveness, as people wanted to live and work in buildings with high ratings. Similar to LEED’s method, environmentalists and the EPA could push for prioritizing green practices by advocating for using AI from developers that are actively trying to find solutions and reduce environmental damage. Even simple initiatives like pushing for AI developers to create energy-efficient algorithms and equipment, or the implementing technology for reusing water in data centres to mitigate consumption would make a difference. 

The government’s action plays a crucial role in managing this problem. The EPA should amend the Clean Air Act to explicitly classify data centers as major sources. The Senate can update the Artificial Intelligence Environmental Impacts Act of 2024 bill to make AI-specific environmental reporting rules mandatory. The government could also motivate AI developers to cooperate with certain incentives by the EPA implementing a tiered reward and punishment system for AI developers. 

AI companies want government support to secure massive infrastructure investments and boost public R&D funding. The EPA can monitor the emissions of major source AI facilities, and developers that can prove that their new implementation of greener technologies has significantly decreased their environmental footprint, through comparing water usage and carbon emissions reports that they will release, will receive financial support. Companies that worsen their environmental footprint or fail to report their emissions and water usage will meet carbon pricing mechanisms. The EPA will decide what the levels look like; an example of how to scale the tiers could be how much percent of emissions were reduced/increased, or it could remain on the same scale as the CAA with counting how many tons of carbon the company emits. E.g. With the assumption that major sources are stationary facilities that emit over 100 tons of carbon each year, if a developer cuts their emissions down by 25 tons, they receive x amount of grants, and if they cut them down by 50 tons they receive x+x amount in grants. However, if their emissions increase by 25 tons per year, they are taxed, and if their emissions increase by 50 tons more per year, the tax is larger. etc. AI is public, but the government can encourage its citizens to use environmentally sound developers. In a sense, the competition with LEED certification could be replicated here, where AI developers compete with each other to reduce their environmental impact. It is possible to pursue innovation without disregarding its cost to the planet. 

VI. Conclusion

Artificial intelligence is arguably the most influential technology of the modern era. Governments, corporations, and individuals increasingly rely on AI systems to enhance productivity, accelerate innovation, and strengthen international standing. Evidently, the infrastructure to develop and operate AI demands significant amounts of electricity and water, contributing to increased carbon emissions and resource consumption. While existing statutes regulate industrial pollution and resource use, they were not enacted with the evolution of AI in mind, creating large gaps in oversight and accountability. AI is only going to continue to expand, and without clearer frameworks, AI developers will continue to disregard environmental costs and benefit from the rapid technological growth.

Addressing these challenges does not have to be so dramatic as to halt AI development, but it does require modernizing existing legal decrees. Implementing firm disclosure requirements and clarifying how existing environmental statutes apply to AI-related infrastructure is nonnegotiable. The rise of artificial intelligence both presents an opportunity and a responsibility. As society integrates AI into everyday life, regulations must judge the sustainability of technological progress. The goal to “secure a brighter future for all Americans” and “enhance America’s global AI dominance in order to promote human flourishing” should include safeguarding the natural resources and Earth’s atmosphere upon which future generations depend.

[1]  Exec. Order No. 14,179, 90 Fed. Reg. 8741 (Jan. 23, 2025).

[2] Li, Pengfei, Jianyi Yang, Mohammad A. Islam, and Shaolei Ren. "Making ai less' thirsty'." Communications of the ACM 68, no. 7 (2025): 54-61.

[3] Samantha Burchard, The true price of every CHATGPT prompt Earth Day (2026), https://www.earthday.org/the-true-price-of-every-chatgpt-prompt/.

[4] Kritin Rajpal & Name, The Boar (2025), https://theboar.org/2025/05/how-much-do-your-chatgpt-prompts-impact-the-planet/.

[5] Samantha Burchard, The true price of every CHATGPT prompt Earth Day (2026), https://www.earthday.org/the-true-price-of-every-chatgpt-prompt/.

[6] Kritin Rajpal & Name, The Boar (2025), https://theboar.org/2025/05/how-much-do-your-chatgpt-prompts-impact-the-planet/.

[7] Samantha Burchard, The true price of every CHATGPT prompt Earth Day (2026), https://www.earthday.org/the-true-price-of-every-chatgpt-prompt/.

[8] Kritin Rajpal & Name, The Boar (2025), https://theboar.org/2025/05/how-much-do-your-chatgpt-prompts-impact-the-planet/.

[9] Sophie McLean, The environmental impact of CHATGPT Earth.Org (2024), https://earth.org/environmental-impact-chatgpt/.

[10] Kritin Rajpal & Name, The Boar (2025), https://theboar.org/2025/05/how-much-do-your-chatgpt-prompts-impact-the-planet/.

[11] Akhand Pratap Singh Chauhan & Srishti Sinha, AI and environment law Lexology (2025)

[12]  Senate - Commerce, Science, and Transportation, Artificial Intelligence Environmental Impacts Act of 2024 (2024). https://www.congress.gov/bill/118th-congress/senate-bill/3732/text.

[13] Summary of the Clean Air Act 42 U.S.C. §7401 et seq. (1970), EPA (2026), https://www.epa.gov/laws-regulations/summary-clean-air-act.

[14] María Montero, From cloud to cup: How much water does your CHATGPT drink? IE Insights (2025), https://www.ie.edu/insights/articles/from-cloud-to-cup-how-much-water-does-your-chatgpt-drink/.

[15] Lawson, Ashley J.; Offutt, Martin C.; Zhu, Ling,, Data Centers and Their Energy Consumption: Frequently Asked Questions (2026), https://www.congress.gov/crs-product/R48646.

[16] LEED rating system, LEED rating system | U.S. Green Building Council, https://www.usgbc.org/leed

[17] Kritin Rajpal & Name, The Boar (2025), https://theboar.org/2025/05/how-much-do-your-chatgpt-prompts-impact-the-planet/.

[18] Exec. Order No. 14,179, 90 Fed. Reg. 8741 (Jan. 23, 2025).