Understanding the Environmental and Community Impact of AI and Data Centers

Asking an AI chatbot a simple question consumes about as much electricity as a microwave does in one second. However, giving the AI a complex task, which it handles independently, involves significantly more energy—equivalent to running a laptop for nearly six hours. This disparity presents a significant environmental consideration, first noted after the widely referenced AI research was released. Beyond electricity, the environmental costs extend to water systems, your power bill, and local tax revenue, sparking debates about AI’s impact.

This article doesn’t argue for or against AI use. Instead, it examines AI’s environmental footprint, its implications, and who makes the decisions. As AI stands ready to revolutionize life and potentially alter fortunes, we aim to help you make informed choices about using AI or considering data center projects in your community.

AI Wins By a Mile?

In February 2024, researchers from three universities published a study in the journal Scientific Reports, revealing that an AI generates 130 to 1,500 times less carbon dioxide than a human writing the same amount of text.

The study calculated that an AI query emits about 2 grams of carbon dioxide, including the energy used to train the model, whereas a human writing the same text produces approximately 1,400 grams. The researchers based their calculations on the average American’s annual carbon footprint of roughly 15 tons, divided by the total hours in a year, and then applied it to the 48 minutes taken to write a page.

However, readers quickly identified issues with this approach. They noted that people contribute to carbon emissions regardless of writing activities, as they still need to heat their homes and drive to stores. The actual change is the energy consumption of the writer’s laptop, which results in about 2 kilograms of CO2 emissions over multiple days, not the 36 kg assumed by the study’s method.

The Rematch, With the Answers Graded

In November 2025, a follow-up study was published in the same journal. Nolan Woo, the author, highlighted that the original study ignored the quality of AI’s work. He conducted a test with graded answers, incorporating the time and energy required to achieve correct responses into the total emissions. Using programming problems from a national high school computing contest, he tested four AI models, feeding errors back into the AI for up to 100 attempts.

The emission results varied based on the AI model:

• The smallest model, when correct, emitted 20% to 59% of the carbon a human programmer would.
• Larger models, like ChatGPT or Claude, produced 5 to 19 times more carbon than a human.
• Failed attempts consumed 8 times more energy than successful ones.

Emissions were influenced more by the number of attempts than by the model’s size or cost. Woo also noted that across four years of contests, AI solved every problem in only one year, often struggling to perform. Nevertheless, recent models have significantly improved their coding accuracy, achieving over 95%, according to Stanford University’s AI Index Report.

Then Agents Changed the Math

Both studies assessed single-question scenarios, which don’t reflect how AI is utilized in 2026. Today’s AI tools plan, search, read, revise, and loop for extended periods, with complex tasks consuming hundreds or thousands of times more energy than text generation.

Researchers at KAIST, South Korea’s leading engineering school, assessed the “hidden energy” cost of AI. They discovered that a single-agent task consumed an average of 348 watt-hours, 136 times the energy of answering a simple question. This discrepancy arises because agentic tasks take much longer and involve multiple AIs handling different aspects of the task, leaving chips idle and still consuming power while awaiting other processes.

Despite this inefficiency, the International Energy Agency noted an unprecedented improvement in AI’s energy efficiency per task, according to an April 2026 report. Reducing waste in the current infrastructure will lower AI’s environmental cost and enhance profitability, despite AI currently being a financial loss overall.

The Break-Even Point

We can pinpoint when AI ceases to be the cleaner option. Consider a research project requiring 20 hours to complete, involving reading 40 sources and writing a 5,000-word report. This task results in about 2 kilograms of carbon from the laptop, monitor, and associated pollution.

Determining how many AI agent requests match this 2-kilogram mark depends on the data center’s power cleanliness.

This last row is significant. Large AI data centers are often constructed in areas where power is less clean than average. A 2026 study indicated that the electricity serving these centers is about 48% dirtier than the national average. Most serious AI research tasks today require far more than 11 requests.

The Catch Nobody Catches

Analyses favoring AI often assume that the machine replaces, rather than supplements, a person’s work hours. However, evidence suggests otherwise, as AI leads to increased workloads.

A National Bureau of Economic Research study, which is still under review, found that people heavily exposed to AI at work log an additional 3.5 hours weekly. Surveys by staffing and consulting firms also indicate that only a quarter of employees use AI time savings for personal time, while most fill it with more work.

If an AI workflow produces three times the output of a person and is used three times as often due to its speed and low cost, it results in a ninefold increase in usage.

This phenomenon aligns with Jevons’ Paradox, identified in 1865 by economist William Stanley Jevons, who observed that more efficient steam engines didn’t decrease coal consumption in Britain; instead, cheaper power led to more applications, increasing coal use for decades.

Water Impacts

Water usage is a common concern in local disputes over data centers, with the smaller figures often cited. According to Lawrence Berkeley National Laboratory, U.S. data centers used around 17.4 billion gallons of water in 2023 for cooling. Although this is less than a day’s irrigation for the nation’s corn crops, Orennia reports that it could double or quadruple by 2028.

The larger, often overlooked figure is the 211 billion gallons of water used to generate the electricity consumed by these data centers, 12 times the cooling figure. Power plants are the largest freshwater consumers in the U.S., and data centers contribute to this through their power bills.

Data centers may switch to air cooling and claim minimal on-site water use, but their actual water footprint remains unchanged. Only data centers powered by clean energy without evaporative cooling can reduce both figures.

Reports from companies are inconsistent. Google reported using 7.7 billion gallons at its data centers globally in 2024, with 1 billion gallons used at its Council Bluffs, Iowa site alone. In Texas, 83% of the state’s 341 data centers had not submitted required water reports.

Industry groups correctly note that total data center water use remains under 1% of the nation’s total. However, the national average offers little solace to a town in the Arizona desert.

The Charge On Your Power Bill

One of the fastest-moving costs from technical detail to household concern is electricity pricing.

Here’s the process: PJM, which operates the power grid for 65 million people across 13 states and Washington, D.C., conducts an annual auction, paying power plants to ensure electricity availability on the hottest and coldest days. These payments are included in customer bills.

The auction price increased from about $29 per unit for 2024-2025 to around $329 per unit for 2026-2027. An independent monitor found that data centers contributed to 63% of this increase, amounting to $9.3 billion collected from customers. The Natural Resources Defense Council predicts an average family in the region will pay about $70 more monthly by 2028.

The impact is evident in utility bills nationwide. Pepco customers in Washington, D.C., faced $21 monthly increases, while home electricity rates in Ohio and Pennsylvania rose by 9% and 14% over the past year.

The industry argues that this is only part of the story. Virginia’s state watchdog agency, JLARC, found that data centers support approximately 74,000 jobs and contribute $9.1 billion annually to the state’s economy. Most of this stems from construction, not operation. Developers also claim that adding a large customer can distribute the grid’s fixed costs across more users, potentially lowering rates. This depends entirely on contract terms.

States are experimenting with managing grids to separate household power and pricing from energy allocated to data centers. Virginia established a separate rate category for data centers, while Ohio approved a rule requiring 12-year contracts and minimum payments for major users. Both aim to ensure new demand funds the power plants it necessitates, rather than spreading costs across everyone.

Why Towns Are Saying No

This situation has led to one of the largest land use conflicts in recent U.S. history. Data Center Watch, which monitors local opposition, reported at least 75 projects worth around $130 billion were blocked or delayed in the first three months of 2026. This matches the total for all of 2025, with active opposition groups more than doubling to 833 across 49 states. Over 300 related bills were introduced in statehouses in the first six weeks of the year.

Opposition to data centers transcends party lines, which is unusual for environmental issues. Republican officials focus on tax breaks and grid strain, while Democrats address water and pollution concerns. However, both ultimately vote similarly.

Two towns illustrate the outcomes of these disputes, each with a different conclusion.

Tucson, Arizona, emerged victorious. Amazon sought a 290-acre campus that could expand to 10 buildings. The project advanced through Pima County under a 2023 confidentiality agreement concealing Amazon’s involvement from the public and some officials. County supervisors approved a land sale in June 2025 without full knowledge, but when residents learned the project’s scale, over 1,000 attended a hearing. In August, the city council voted unanimously to halt the project, refusing land annexation or city water supply. The developer considered it a loss of significant tax revenue and jobs.

The developer later purchased the land under a different company name, moving the project just outside city limits, where the council’s decision didn’t apply. Although the organizing succeeded, the city’s authority to block the project ended at the boundary.

Saline Township, Michigan, faced a more challenging situation. In September 2025, both the planning commission and township board opposed rezoning 575 acres of farmland for a $16 billion data center linked to OpenAI and Oracle. Two days later, the developer sued the township, arguing that the township’s lack of industrial zoning effectively banned an allowable use.

With roughly 2,900 residents, the township couldn’t afford a prolonged legal battle and settled within weeks. Construction began in November, with residents negotiating about $14 million in community benefits, including fire department funding and farmland protection. Only one township supervisor voted against the settlement.

Both disputes are now occurring on a larger scale. In Virginia, a coalition of data center companies, chambers of commerce, and construction unions spent months defending a multibillion-dollar tax incentive through a TV ad campaign. Nationally, a political group called Leading the Future launched with $140 million to support pro-AI candidates, including $50 million each from a venture capital firm and OpenAI’s president. AI company employees formed the Guardrails Alliance, raising small donations to counter such efforts.

New York recently became the first state to halt new large data center construction. At least 19 Michigan towns enacted their own building moratoriums after Saline’s experience.

What You Can Do

While individual choices matter less than infrastructure decisions, your AI usage will ultimately influence the extent of data center infrastructure.

• Match the tool to the job. A simple chatbot question results in a few grams of carbon, about what a car emits in 2 seconds. In contrast, the same question can cost 19 kilograms of carbon, equivalent to 47 miles of driving, if handled by an agent in multiple steps.
• Stop the loops. Failed and repeated attempts significantly increase emissions. If a tool is stuck, stopping it is both practical and environmentally friendly.
• Smaller is not always better. A small model that fails 20 times can produce more carbon than a large model that succeeds on the first try.
• Read your power bill. Auction costs typically appear as a separate line item, often called a capacity charge. If you reside in one of the 13 states, this line reflects data center demand reaching you.
• Go to the utility hearing, not just the zoning meeting. State utility commissions decide who bears the cost of new power plants, and they accept public comments. This is where the cost burden is determined.
• Ask what confidentiality agreements cover. Several disputes arose from residents discovering that officials had signed away the right to disclose the customer’s identity. Any resident can inquire about this at a public meeting.
• Ask who is counting, and how. When encountering claims that AI is far cleaner than humans, verify whether the calculations include the person’s commute and household emissions. This one choice dramatically alters the outcome.

Your Next Choice Matters

For straightforward tasks that an AI accomplishes quickly, it indeed presents a lower environmental impact. However, for in-depth research requiring an agent to loop, search, and make corrections, humans currently have a cleaner footprint, with the gap widening with each additional step.

The key takeaway is that the energy consumed per question isn’t the most critical factor. Instead, where data centers are built, which utilities serve them, whose water they use, and who signs the power contracts will significantly influence AI’s environmental cost. These are public decisions made in forums open to public input, and they are being decided rapidly.