By Isabella Aranda Garcia
Artwork by Molly McGarvey
You’ll see the innovative and amazing things AI can do as you read through this issue. Whether you’re a skeptic or have been all-in since day one, there’s a tremendous amount of nuance behind the scenes that most people are not privy to. This new technology is impacting communities in a myriad of ways, ranging from the cost of building data centers to the amount of natural resources these centers consume and their life-changing consequences, including incessant noise pollution and an increase in consumer energy bills.
To better understand data centers and how they impact our environment, we spoke to several experts who revealed what AI companies don’t willingly share with the public. Here’s what you should know before a data center pops up in your backyard.
Let’s start with the positive
AI tools are being created and used for sustainability efforts, from climate mapping and smart-grid management to precision agriculture, risk analysis, dependency networks and emergency services resource allocation. Since environmental data can be complex and variables like weather, air quality, water conditions and wildfire activity are updated hourly or in near real time, AI can make it easier to analyze this data, helping experts make informed decisions.
Manny Patole is an industry assistant professor at The Center for Urban Science + Progress (CUSP) at the NYU Tandon School of Engineering who leads a research initiative called “Weight of the Cloud” that examines the environmental, social and governance (ESG) impacts of AI and data center infrastructures. He says AI tools have the potential to improve the environment—both natural resources and the physical, lived-in spaces of communities—on a local level, especially with respect to resource optimization. Within the field of Property Technology (PropTech), AI-based control algorithms are already being used to reduce unnecessary heating and cooling by predicting energy demand and optimizing HVAC systems. Additionally, AI is being used to analyze traffic flows and ridership patterns to help planners identify promising locations for new public transit infrastructure, potentially improving service, increasing transit use and reducing reliance on private automobiles.
The potential is similar when it comes to natural disasters. In a 2023 article in Computational Urban Science, researchers trained machine learning models to map hurricane evacuation patterns based on smartphone location data. Tools like this could help municipal emergency planners anticipate how to position emergency food, water and medical resources where they will most likely be needed before a hurricane.
But it all comes at a cost. “When you look at the sustainability side, you’re mitigating the harm and then examining how to leverage AI to resolve some of these ecological challenges,” Patole explains. “But there’s always a chicken and egg; if we didn’t use it, then we wouldn’t have these problems in the first place, right?”
The major public concern
Currently, data centers are mostly powered by non-renewable energy, and there are conflicting opinions about whether data centers could be run sustainably. “The scale they’re operating at is difficult to conceive,” says Marc Conte, Ph.D., an environmental economist and professor in the economics department at Fordham University. As an example, he points out that Meta’s Hyperion data center in northern Louisiana is roughly one-quarter the size of Manhattan, and its energy demand is equal to that of hundreds of thousands of single-family homes. “So all of a sudden, you’re adding an individual structure that might double the demand for electricity that the utility company is used to providing, which is a huge challenge for utilities.”
Conte also brings up the speed at which these AI companies want to build their data centers and become operational. “If you have a new power-generating unit, and you want to connect it to the grid, you’ll have to wait years to get it online,” he explains. “But these AI companies are building facilities in 12 to 24 months, and they want their new facilities to come online immediately. There’s no way they can wait two years for the new energy these facilities need to meet the demand it takes to power them. So what are they doing? They’re building their own power plants, and typically the plants they’re building are not powered by renewable energy.” Instead, these power plants often run on fossil fuels or natural gas.
Water use is also a concern, Conte says, pointing to Meta’s data center in Newton County, Georgia. The county relies on rainfall runoff collected in Lake Varner, and without infrastructure upgrades, it could face a water-supply deficit by 2030—a risk underscored on April 27 when the county reported extreme-drought conditions. To combat this problem, many AI companies, like Google and Microsoft, are building closed-loop systems, which move the water used as a coolant through sealed pipes where it absorbs heat and recirculates to reduce internal water consumption. But even this system is not sustainable enough, because as the water gets hot from use, it requires additional energy to cool it down, making it more of a trade-off rather than a true clean energy solution.
The real cost and who pays the price
The price to use most LLMs is free to an extent; to have access to all the features of an AI model, you will need to purchase a premium version. This, according to Conte, creates an economic market failure where prices don’t reflect the true cost of an item. “People are using these things a lot more than they would if they had to pay the true costs,” he says.
So, who’s footing the bill? We are, says Conte; we just don’t get an itemized receipt. In a recent New Jersey election, a state with 48 data centers, candidates spoke on the cost of power because voters frequently raised it as a concern. Over the past year, the price has risen by roughly 20 percent. Conte believes the price spike could be due to PJM Interconnection, the largest regional transmission organization in the United States, operating across all or parts of 13 states and Washington, D.C., including New Jersey. Politicians in the state have faulted PJM as well. However, Conte says it’s more nuanced than the electricity company simply raising rates. He points to a 2025 Bloomberg study, which he believes demonstrates that “parts of the grid that saw the largest increase in electricity prices were closer to existing data centers, compared to those that were further away.”
And the bills keep growing. When companies seek a location for a new data center, states compete for the tax revenue, and utility companies hope the data centers will increase their profits. “There are some concerns about transparency because state officials and utilities conduct the negotiations in secret,” Conte explains. “Effectively, state officials might try to sweeten the deal to attract these data centers, and that could reduce their potential tax revenue.”
Those tax breaks could shift costs or constrain public spending for the community. “If a million is taken out, that million has to be filled somewhere,” says Patole. “To cover it, municipalities must cut public services, raise local fees, use reserve funds or rely on state and federal aid. How these gaps are filled depends entirely on local institutional and political history as well as what trade-offs constituents are willing to make in the short term for the long term. It requires them to consider company promises against actual outcomes.” To balance tax breaks, AI companies promise new jobs, but according to an August 2025 article in Data Center Knowledge, the number of jobs doesn’t actually balance out the cost of these facilities. “There are jobs created, but AI companies often bring in contractors from other places,” Patole explains. “There might be a couple of local hires, but not to the extent that it will offset a million-dollar tax bill.”
Conte points to a ripple effect beyond the data centers. “Because they’re operating at such large scales and because they want things to happen so quickly, they’re pulling construction crews from other projects,” he says. This could result in delays for a community’s future infrastructure projects and small projects, like installing HVAC systems for office buildings, driving up labor costs as skilled workers grow scarcer.
The “true costs” are also reflected in carbon dioxide emissions and long-term consequences, like increased mortality from heat, extreme weather and more frequent natural disasters. “Our estimate of that cost, which is known as the social cost of carbon, is roughly $190 per metric ton of carbon dioxide,” says Conte. In other words, every extra ton of carbon dioxide has a price tag; it’s just not one you can see on your screen when you submit your prompt.
Can data centers reduce their carbon footprint?
The ideal solution to reduce data centers’ energy consumption is to power them with renewable energy. However, that might not be practically possible. Renewable resources like solar and wind power require a lot of land, especially given the scale of energy required for data centers. Conte offers nuclear as a solution, specifically small modular reactors (SMRs), which Microsoft plans to utilize on Three Mile Island, a nuclear power plant in Pennsylvania. Under a 20-year power purchase agreement with Constellation Energy, Microsoft is funding the restart of the plant’s Unit 1 (rebranded the Crane Clean Energy Center), aiming to bring 835 megawatts of carbon-free electricity onto the PJM grid by 2027. “They will build small nuclear plants adjacent to the data center and generate their electricity that way. As an environmental economist, I think nuclear will need to play a big role if we are going to avoid catastrophic climate change.”
But scaling that concept is challenging. The U.S. Department of Energy’s 2024 Data Center Energy Usage Report projected that all data centers, including those that power more than AI, will use 325 to 580 terawatt-hours (TWh) of energy in 2028. By comparison, an 835 MW nuclear unit used by Microsoft will generate 6.6 TWh annually. That means the overall electricity demand from U.S. data centers will be roughly 49 to 88 times Crane’s annual output. So, for all U.S. data centers to utilize nuclear energy, AI companies would need 50 to 90 units similar in size to what Crane is building. These units would take approximately 10 or 11 years to build, beginning with a five-year application review from the Nuclear Regulatory Commission, followed by another five years (or more) of construction. The next big hurdles are training and retaining a large specialized workforce, expanding factories and fuel services, upgrading transmission lines and coordinating many state and local approvals.
Melissa Hoffmann, a Westchester-based organizer for Food & Water Watch, isn’t sure this concept is feasible. “It takes a really long time to select a location for a nuclear plant because of zoning restrictions, permitting and community support and/or backlash, so we won’t see them power data centers anytime soon,” she explains. “Additionally, they still rely on uranium, which is a health risk for the communities they are in, and it takes multiple SMRs to meet the power output of a single conventional reactor, increasing the risks of nuclear disaster. Plus, they’re super expensive to build and maintain, and the costs are paid by ratepayers, like electricity customers. I see it as a false solution.”
Hoffmann is also concerned that data center development will slow the transition to renewable energy, even if the data centers are powered by renewable energy. “We need to focus on reducing the use of fossil fuels as rapidly as possible,” she says. “And we need to prioritize having a reliable grid that does not disadvantage or fail certain communities that live near data centers, whether through spikes in energy demand, brownouts or increased utility burdens.” But even if our grid ran on renewable energy, Hoffmann believes data centers’ energy consumption would still be a major concern.
Since transitioning to green energy is a longer-term solution, Patole is currently exploring ways to reduce economic and environmental damage. One solution is for organizations to switch to small language models (SLMs). “The New York City Council is leveraging concepts of small language models,” he explains. “These are locally hosted, which means they’re utilized within their own organization or institution and only focus on the work they’re doing.”
Patole envisions municipalities building their own SLMs for residents to use. Because SLMs have fewer parameters, they require less computation per query and less cooling, meaningfully reducing their energy and water footprint compared to LLMs. “They wouldn’t have to go out into the internet world and pull information, which is where a lot of energy is theoretically expended,” says Patole. Local, smaller-scale fixes like SLMs are currently the exception, not the rule.
The push to build fast
Even with greener options on the table, there are major concerns about how AI infrastructures are currently being facilitated. The federal government has entered the global AI race, hoping to house the companies that will continually produce the most innovative and cutting-edge AI tech. But to achieve this, it’s cutting corners and creating paths for companies to build data centers swiftly, causing a lack of transparency about how many resources are being used, the hidden costs and the lack of regulations (see pg. 44 for more on AI regulations).
In the 2025 America’s AI Action Plan, the White House identifies data centers and the energy infrastructure that powers them as necessary for U.S. AI leadership. The plan explicitly champions faster permitting and recommends the federal government explore creating a new nationwide Clean Water Act Section 404 permit exclusively for data centers. This permit would exclude them from submitting the standard pre-construction notification that allows the U.S. Army Corps of Engineers to determine if the project qualifies for a general permit and what conditions, mitigation or other approvals might be required. That same day, July 23, Trump issued Executive Order 14318, which accelerates federal permitting of data center infrastructure. The order directs federal agencies to ease regulatory burdens and expedite environmental review and permitting for qualifying AI data centers and supporting infrastructure projects.
Currently, no policies have been adopted to offset the energy costs for these data centers, and their rapid construction could impact communities if they are not properly planned. But, on July 14, 2026, New York Gov. Hochul signed a moratorium on data centers that consume at least 50 MW, pausing their development statewide for one year. Currently, there are 20 operational and 35 proposed data centers throughout New York, including three operational and two proposed in Westchester. The proposed facilities range in size from about 50 to 2,800 MW of electricity demand. The state has not released a complete public list detailing which of these projects had Department of Environmental Conservation (DEC) permit applications that were put on hold.
This moratorium directs the Department of Public Service to consider and develop mechanisms that would protect ordinary utility customers from data center-driven grid costs. It also will allow time to prepare a statewide Generic Environmental Impact Statement (GEIS) that will assess data centers’ potential effects on energy demand, water use and quality, air quality and noise levels, as well as the disproportionate impacts on disadvantaged communities. Their findings will guide a subsequent review so the state can finalize standards before lifting the moratorium.
Hochul’s moratorium is the first in the country; other states are now considering similar measures. Patole believes this pause offers us, the people potentially impacted, a chance to pay attention to what is happening behind closed doors and have a say in whether we opt for this new technology and create better rules to avoid the negative impacts. “I think that’s what’s happening now with this moratorium—it’s letting us take a breath,” says Patole. “Let’s figure out what’s going on. Let’s understand the local economic impacts, the social impacts, the policy and governance impacts, as well as the environmental impacts.”
Hoffmann hopes more local politicians follow Hochul’s lead. “Any town or county can pass a local moratorium on data centers to study and develop regulations,” she explains. Patole agrees. “You don’t want to hang a decision on a city or a community that forces them to pay it back for the next 20 to 40 years based on something that was made in haste because an AI company didn’t want to be overlooked,” he points out. “That’s the idea of FOMO, right? These companies will decide and act less drastically if they have a little bit more time and if we implement robust guardrails designed to protect all of our constituents—both human and non-human alike.”
On a federal level, there are two moratorium proposals in committee—one in the House and one in the Senate. Hoffmann encourages readers to call their representatives to co-sponsor the proposals (as of deadline, neither of our senators nor local representatives are sponsors).
The experts believe that politics can, and should, stay out of this issue. “It’s kind of a grassroots movement, which in this time of extreme partisanship, is actually not a partisan issue,” Conte notes. “We’re seeing people reject data centers in very red communities and in very blue communities. People are worried about the prices these data centers will have on their everyday lives.”
Going forward
What can New Yorkers do during this one-year pause to help mitigate the potential harms of data centers? A major focus should be placed on community and urban planning. Patole visits communities impacted by data centers and speaks to residents, hoping to provide some transparency. “Participatory governance is a beautiful ideal, but everyday survival gets in the way,” he says. “They’re working, raising kids and maintaining a household—when these things pop up, it can be very frightening. In my project, and when working with some of my students, we want to determine how we can provide this information in a way that’s accessible.” But the impacts are hard to quantify because, as a 2025 report in the scientific journal Patterns points out, data center operators don’t distinguish between AI and non-AI workloads in their environmental reports.
Another part of the problem is where data centers are located. “These data centers might actually be closer to urban areas than we think,” says Patole. “We know that time and time again, BIPOC communities—the ones who are underrepresented—often experience the greatest environmental impact. There are healthcare concerns that may not present themselves until maybe three, five or 10 years down the road,” he says, adding to why he wants standardized definitions of environmental and community harm before any moratorium lifts.
New York State’s Responsible Data Center Development Act, proposed in June 2026, takes a similar approach on a larger scale. It would require data centers with at least 20 megawatts of peak demand to bear, through separate utility service classifications, the costs of infrastructure upgrades, administration, improvements or additions, operations and certain electricity- and water-related price increases related to serving those facilities. It would also require new large data centers—and existing ones undergoing a major expansion adding at least 20 MW—to fund host-community benefit and impact-mitigation programs. Possibilities include drinking water infrastructure, broadband improvements, household energy technologies and pollution mitigations. And, if implemented, it would require at least one in-person public hearing in the host communities at least three months before the state grants the permit, but it does not give the community a chance to veto the project.
With the moratorium and proposed law, it may seem that New York is acting in the public interest, but that ultimately depends on the follow-through. Patole says you have more of a say than you think, and you should use it. “It’s the power of one, but it must be built on sharing fact-based, verified knowledge within and between communities.” Hoffmann echoes this sentiment. “Show up to your town board meetings regularly and ask if any data center company or other major development project has reached out and is considering building.” If you’re reading this in September, there are about 10 months left on Hochul’s moratorium, and our experts have outlined the path they hope you will take. The rest is up to you.
This article was edited by Julie Schwietert Collazo and fact-checked by Gia Miller. The artist used Adobe Creative Suite.
This article was published in the September/October 2026 edition of Connect to Northern Westchester.
Isabella Aranda is a designer, writer and social media specialist with an M.A. in emerging media from the New Media Institute at the Grady College of Journalism & Mass Communication. Driven by curiosity and inspired by timeless modern design, she blends creativity and strategy to craft compelling narratives that engage diverse audiences.
Her expertise spans digital marketing, content creation and UX design, with notable achievements such as co-creating the Georgia On Your Mind podcast and leading digital campaigns that significantly boosted engagement. A Venezuelan immigrant, Isabella brings a multicultural perspective to her work, enhancing her ability to connect with and inspire others.
Molly McGarvey
Molly is a graphic designer & photographer currently designing for CTNW and assisting with the secret behind-the-scenes magic. A recent Purchase graduate with a BFA in graphic design, Molly is passionate about design for change, with a strong interest in environmentalism. In her free time, she enjoys walks, hikes, riding four-wheelers, cooking and watching true crime documentaries with her cat.
