Most scientists believe that climate change threatens mankind’s future and that human activity contributes mightily to that change, especially as it involves fossil fuels. A November 2024 update on climate progress found much reason for pessimism, particularly because fossil fuel subsidies are at an all-time high and funding for fossil fuel-prolonging projects quadrupled from 2021 to 2022. President Trump’s energy policies are designed to exacerbate this situation.

Taking a utilitarian approach to the matter, many people feel a moral obligation to replace their gas guzzlers with electric vehicles (EVs) in order to slow the oncoming climate disaster. But all sides of the equation must be weighed. Critics point out, for example, that EVs are charged with electricity generated by polluting power plants, that manufacture of their batteries creates pollution, and that batteries must be replaced frequently. Full consideration of all the evidence makes the utilitarian moral argument closer than many EV advocates admit.

Whether AI development will harm or help the environment is also a critical issue today, and that is the subject of this post. In earlier blog posts, we have highlighted the contentious and rapidly changing factual environment surrounding AI evolution, emphasizing that the stakes are high, but the factual certainty is low. The fluidity of the situation is highlighted by such recent developments as: announcement of Open AI’s Operator, the surprise rollout of China’s DeepSeek in late January 2025, and the mid-February 2025 revelation that Microsoft has created a new state of matter that advances the development of quantum computers which could greatly empower artificial intelligence. In this post, we focus narrowly on the environmental plusses and minuses of AI development, finding the same bottom line: the factual situation is changing with lightning speed and drawing any firm utilitarian conclusions is difficult.

AI’s impact on emissions, energy use, and water consumption

Emissions

Journalist Robert Wright has observed that once upon a time, “the most common political message you heard in Silicon Valley was about the need to fight climate change. But that was back when huge power consumption was something old-fashioned Rust Belt industries did, whereas software was a ‘clean’ technology. Things have changed.” They have changed because of AI.

At the dawn of 2025, there were almost 3,000 data centers in the U.S. Carbon emissions from data centers tripled between 2018 and 2025, and these emissions were significant– in the year preceding August 2024 data centers caused 105 million metric tons of CO2, roughly 2.18% of total US emissions. Google’s annual emissions rose 50% between 2019 and 2024, mostly due to AI. It stopped claiming to be carbon neutral.

A significant complication is that 95% of these data centers are located in places like West Virginia where the sources of power are coal and other fossil fuels so that their “carbon intensity” is 48% higher than the national average. The inconsistency of solar and wind power makes them less desirable choices to power data centers. ChatGPT and other AI tools are always on.

Energy Usage

It will be difficult to pare back the energy usage that causes these emissions and simultaneously drive the AI revolution forward. It takes huge amounts of electricity to train AI tools such as Chat GPT, and the more sophisticated the tools become, the more energy intensive they are, at least so far. For example, training GPT-4 took 300 times more energy (60 million kWh) than training GPT-3, and GPT-5 “is expected to require considerably more energy than that, perhaps ten or even a hundred times as much, and dozens of companies are vying to build similar models, which they may need to retrain regularly.” (Marcus)

Once in operation, these tools continue to gulp energy. By some estimates, a typical request to ChatGPT takes 10 times the electricity (10 kilojoules) of a Google search. And ChatGPT has nearly 200 million users. “If ChatGPT were integrated into the 9 billion searches done each day, the IEA [International Energy Agency] says, the electricity demand would increase by 10 terawatt-hours a year—the amount consumed by about 1.5 million European Union residents.” (Calvert)

All told, data centers account for 4.59% of all energy usage in the U.S. (double that in 2018). A recent report estimates that U.S. data centers went from 1.9% of total electrical consumption in 2018 to 4.4% in 2023 and will consume 6.7% to 12% in 2028.” (Klein)

Given these projections, coal plants which otherwise would have closed due to pollution concerns are staying open. And many leading AI companies (Meta, Microsoft, Amazon, etc.) are seriously considering tapping nuclear energy in the future. Nuclear power, of course, carries its own environmental risks.

Water

As these data centers enable training and usage of AI tools, another serious environmental concern arises–a shortage of fresh water. As Karen Hao recently reported in The Atlantic, “global AI demand could cause data centers to suck up 1.1 trillion to 1.7 trillion gallons of fresh water by 2027.” This water is used to dissipate heat, both as the AI program is trained (training a large language model [LLM] like Chat GPT-3 can take millions of liters of fresh water), and then as it operates (running GPT-3 inference for 10-50 queries uses more than 2 cups of water, and GPT-4 uses even more).

And we’re not even counting the fact that it takes approximately 2,200 gallons of Ultra-Pure Water to produce a microchip.

AI companies are trying to do better

 
Realizing the adverse impact that AI development is having on the environment, many leading AI firms are trying to do better.

Many big firms are adopting carbon offset plans to make up for their carbon emissions. In 2024, as just one example, Microsoft committed to buy enough renewable energy from Brookfield Asset Management from 2026 to 2030 to power its AI operations with carbon-free energy. This is “better than nothing,” says Fengqi You of Cornell, “but it’s definitely not an ultimate solution.” Robert Wright observes: “The big AI companies like to finesse [the climate change issue] by saying they’re trying to use green energy sources and, when they can’t will compensate by helping to finance future green energy projects. But the compensation is only partial, and most green energy sources they use in the nearer term are ones that someone else would have used if they hadn’t.”

Although the big AI firms are talking wind, solar, geothermal, traditional nuclear, and fusion, we should remember that natural gas, which accounts for 42% of electricity generation but releases much methane (an atmospheric heat trapper) as well as carbon dioxide, is the likeliest source of big increases in power. Jason Bordoff, the founding director of the Center on Global Energy Policy at Columbia University says that in the short-term, natural gas is the obvious source of a big expansion in electricity production.

Still, some firms are trying. Data centers are developing more efficient cooling technologies, researchers are designing specialized hardware such as new accelerators, and Nvidia claims it has a new chip that can deliver 30X the performance while cutting energy use by 25%. A research team at BitEnergy AI has developed a new algorithm (“L-Mul”) with the potential to reduce the amount of energy AI uses by 95%, which would obviously be huge. Unfortunately, as this blog is written, the algorithm requires specialized hardware that is not widely available. However, China’s DeepSeek makes it appear that such improvements are possible. Firms have an economic incentive to develop such innovations, but also an arguable moral obligation.

AI may solve our environmental and energy problems

 
No matter how detrimental to the environment the drive to develop AI appears to be, its supporters believe that advances toward artificial general intelligence and even superintelligence just might solve our environmental and energy concerns and therefore there is a moral mandate to push ahead researching and developing AI.

AI tools are already solving complex problems in many areas of human endeavor, and they soon will be much, much smarter. Thus, the late Henry Kissinger and co-authors claim that “we would be wise to explore [the] opportunity that AI provides to solve our climate problems.” Futurist Ray Kurzweil agrees, believing that the future is in clean energy, especially solar, and will progress “exponentially as further cost reductions are driven by AI applied to material discovery and device design. … Continuing exponential gains will also be enabled by convergent advances in materials science, robotic manufacturing, efficient shipping, and energy transmission.” AI has already been put to work modeling and understanding climate events.

Maybe AI will not be so helpful

 
Although AI developers have made progress on many fronts and make more every day, so far they’ve not created a product that has made a meaningful dent in our climate crisis or much of anything else. They very well might in the future, but at the moment this is speculative. Gernot Wagner, a climate economist at Columbia Business School opines: “[AI development is] a massive increase [in emissions] that may not be justified by the productivity gains from AI.” (quoted in Coleman).  Science reporter David Gelles agrees: “Global warming is one of the biggest challenges we face as a species, and the promise of a new technology magically making an existential problem evaporate holds deep appeal. So far, however, there is scant evidence that A.I. will deliver a miracle fix. Incremental gains are possible.”

Conclusion

 
We agree with research scientist Jesse Dodge who argues that AI firms should establish ethical principles designed to avoid climate harm and to reduce it, saying “[i]t needs to be part of the value system.” Garnett suggests AI system designers should:

  • Design energy-efficient algorithms that use minimal computing power
  • Optimize and minimize data processing needs
  • Choose hardware with maximum power efficiency
  • Use data centers powered by renewable energy sources
  • Comprehensively assess the carbon footprint of an AI model
  • Support or engage in research on sustainable AI

Looking at the big picture, it appears to us—and again we stress how speculative our conclusions must be in light of the changing factual environment—that the full-on charge toward rapid development of AI will definitely leave the environment worse off in the short-run, but AI-breakthroughs could create a much better situation in the long-run. How does a good consequentialist weigh those two tentative considerations to reach a moral conclusion regarding further development of AI?

 

Sources:

Brian Calvert, “AI Already Uses as Much Energy as a Small Country. It’s Only the Beginning,” VOX, March 28, 2024.

Climate Action Tracker, Warming Projections Global Update, Nov. 2024, at i.

Jude Coleman, “AI’s Climate Impact Goes beyond Its Emissions,” Scientific American, Dec. 7, 2023.

Thomas Friedman, “Trump Is Going Woke,” New York Times, Jan. 28, 2025.

David Gelles, “The A.I. Climate Dilemma,” New York Times, Oct. 22, 2024.

David Gelles, “What Will Power the A.I. Revolution,” New York Times, January 7, 2025, at https://www.nytimes.com/2025/01/07/climate/artificial-intelligence-power-emissions.html.

Erin Griffith, “They Invested Billions. Then the A.I. Script Got Flipped,” New York Times, Jan. 29, 2025

Taiba Jafari et al., “Projecting the Electricity Demand Growth of Generative AI Large Language Models in the US,” Columbia Center on Global Energy Policy, July 17, 2024, at https://www.energypolicy.columbia.edu/projecting-the-electricity-demand-growth-of-generative-ai-large-language-models-in-the-us/.

Eleni Kemene et al., “AI and Energy: Will AI Help Reduce Emissions or Increase Demand,” World Economic Forum, July 22, 2024, at https://www.weforum.org/stories/2024/07/generative-ai-energy-emissions/.

Saijel Kishan & Josh Saul, “AI Needs So Much Power That Old Coal Plants are Sticking Around,” Bloomberg, Jan. 25, 2024, at https://www.bloomberg.com/news/articles/2024-01-25/ai-needs-so-much-power-that-old-coal-plants-are-sticking-around.

Henry Kissinger, Craig Mundie & Eric Schmidt, Genesis: Artificial Intelligence, Hope, and the Human Spirit 174 (2024).

Ezra Klein, “’Now is the Time of Monsters,’” Washington Post, Jan. 12, 2025.

Ray Kurzweil, The Singularity is Nearer 173, 175 (2024).

Gary Marcus, Taming Silicon Valley: How We Can Assure that AI Works for Us 69 (2024).

Cade Metz, “OpenAI Unveils A.I. Agent that Can Use Websites on Its Own,” New York Times, Jan. 23, 2025.

Cade Metz, “Microsoft Says It Has Created a New State of Matter to Power Quantum Computers,” New York Times, Feb. 19, 2025.

James O’Donnell, “AI’s Search for More Energy Is Growing More Urgent,” MIT Technology Review, Dec. 17, 2024, at https://www.technologyreview.com/2024/12/17/1108894/ais-search-for-more-energy-is-growing-more-urgent/

James O’Donnell, “AI’s Emissions Are About to Skyrocket Even Further,” MIT Technology Review, Dec. 13, 2024, at https://www.technologyreview.com/2024/12/13/1108719/ais-emissions-are-about-to-skyrocket-even-further/ (citing paper from teams at the Harvard T.H. Chan School of Public Health and UCLA Fielding School of Public Health).

Shaolei Ren, “How Much Water Does AI Consume? The Public Deserves to Know,” The AI Wonk (blog), Nov. 30, 2023, at https://oecd.ai/en/wonk/how-much-water-does-ai-consume.

Reece Rogers, “AI’s Energy Demands Are Out of Control. Welcome to the Internet’s Hyper-Consumption Era,” Wired, July 11, 2024.

Aman Tripathi, “AI May Not Need Nuclear Power Feeding, New Algorithm Slashes Energy Use by 95%,” Interesting Engineering, Oct. 10, 2024, at https://interestingengineering.com/innovation/l-mul-algorithm-ai-energy-consumption.

Robert Wright, “Sam Altman’s Imperial Reach,” Washington Post, Oct. 7, 2024, at https://www.washingtonpost.com/opinions/2024/10/07/sam-altman-ai-power-danger/ .

 

Blog Posts

“Artificial Intelligence, Democracy, and Danger”:  https://ethicsunwrapped.utexas.edu/artificial-intelligence-democracy-and-danger.

“AI Ethics: ‘Just the Facts, Ma’am,’”: https://ethicsunwrapped.utexas.edu/ai-ethics-just-the-facts-maam.

“Ethical AI: Moral Judgments Swamped by Competitive Forces”: https://ethicsunwrapped.utexas.edu/ethical-ai-moral-judgments-swamped-by-competitive-forces.

AI Ethics: “Techno-Optimist or AI Doomer? Consequentialism and the Ethics of AI”: https://ethicsunwrapped.utexas.edu/techno-optimist-or-ai-doomer-consequentialism-and-the-ethics-of-ai.