4 Reasons the Southeast Should Pause the Data Center Gold Rush

What’s at Stake:

• Community costs: Data centers lock in unwanted noise and air pollution, drain tens of millions of gallons of water a day, and are approved under agreements lacking transparency, often before residents know a project is coming.
• State & regional costs: Utilities profit from the buildout and pass the costs on to customers whether power demand materializes or not, locking in decades of new fossil fuel power plants. Bills in parts of the region could rise by up to 57% by 2030, even though the odds of these load forecasts materializing are low.
• Economic sector risks: The AI buildout needs about $2 trillion a year in new revenue to pay off — money that can only come from cutting jobs or from debt that could trigger a financial crisis.
• Existential AI misalignment: The same companies are asking the Southeast to bear multiple existential risks while having told regulators they aren’t fully sure they can control what they’re building.

The ask: An 18-month pause, not a ban, to proactively define rules on transparency, pollution, cost, and risk before the concrete is poured.

Silicon Valley has a motto, coined by Mark Zuckerberg in the early days of Facebook: move fast and break things. That mindset is now rolling across the country as a data center gold rush, and it is redrawing the American landscape. Northern Virginia first became the world’s largest data center market, its “Data Center Alley” pulling in so much construction that the power grid could not keep up. Texas came next and is now on track to overtake Virginia by 2030. 

Now the wave is pushing into the rest of the Southeast, and Georgia is already among the hardest hit: metro Atlanta now carries the largest data center construction pipeline in the country. SACE’s five regional focus states — Tennessee, North Carolina, South Carolina, Georgia, and Florida — sit squarely in the path of this invasion. The South already accounts for 48 percent of all planned U.S. data centers.

“Move fast and break things” may be how the Tech Bros think. It is not how thoughtful people in the Southeast think about our land, our air, our families, and our communities. Today the Southern Alliance for Clean Energy is calling for an 18-month moratorium on new large data centers across all five states. A pause is not a no, at least not yet. It is time to get the facts, protect ratepayers, our communities and our Earth’s climate, and put real guardrails in place. Underneath this rushed buildout is a bet that may never pay off, and the Southeast may be left holding the bill no matter how it breaks.

The Buildout Is Outrunning Us

The scale is hard to overstate. Globally, data center power demand is doubling from about 415 to 945 terawatt-hours by 2030, growing four times faster than any other electricity sector. In the United States, Lawrence Berkeley National Laboratory clocked data centers at 176 TWh, or 4.4 percent of all electricity, in 2023, with a path to as much as 12 percent by 2028, and a Department of Energy lab study projects U.S. data center and crypto demand could reach roughly 1,040 TWh by 2030 in its high case. To translate the units: a terawatt-hour is one billion kilowatt-hours, so those 176 terawatt-hours are roughly the yearly electricity of 17 million American homes. And a gigawatt, the unit the regional figures below come in, is about the output of one large nuclear reactor, so a call for nine or ten gigawatts of new gas, like Duke Energy is proposing in the Carolinas, means something close to nine or ten new large power plants. In our region, the impact is more dramatic:

• TVA. Data centers have gone from 1% of TVA’s industrial load in 2019 to 19% last year, expected to double by 2030, with roughly $50 billion in new capital projects, mostly generation, to serve it.
• Georgia. Southern Company’s contracted large load has hit 17 GW, 15 GW of which are in Georgia, and Georgia Power’s roughly 10,000 MW procurement request rests on an improbable 1-in-500 scenario.
• The Carolinas. Duke says data centers drive more than 85 percent of expected load growth and is using those forecasts to justify 9.7 GW of new gas plants. That’s adding a quarter of the utility’s current capacity.
• And the forecasts are speculative. An independent study commissioned by SELC found that utility data center forecasts across the region have roughly a 0.2 percent chance of materializing. The region is being asked to build irreversible fossil infrastructure against demand that is statistically unlikely to appear.

The case for slowing down comes down to four things, and most people arrive at them in the same order. It begins close to home in our communities, with what a data center does to the place you live: the land it takes, the air, the water, the round-the-clock noise, and the secrecy that keeps you from having a say in your own community. 

Next, it widens to our states and regions, impacting your power bill, state regulatory bodies like Public Service Commissions, and the region’s air and water we all share, as utilities lock in decades of fossil fuel infrastructure to serve it as the environmental impacts and bill impacts grow. 

From there it becomes a question about the whole sectors of the economy: speculative debt structures and laying off paid human workers for AI agents.  These are enormous gambles with our economic well-being. And underneath all of it sits the hardest question of all: why are we racing to build a technology before we fully understand its full impacts?

Massive existential questions Silicon Valley Tech Bros debate constantly, things like “AI alignment”, “recursive superintelligence” and a cocktail conversation starter “what is your P-Doom?” Translated for us less enlightened souls: the probability of human extinction by AI. Spoiler alert — for most Tech Bros it’s low but not zero. The rest of us are rarely asked for our opinion about any of this. More fundamentally, do we support racing to build an optional technology that we and they do not fully understand? 

Any one of four levels of concern — community, states and regions, sectors of the economy, and existential questions — is reason enough to pause. Together they represent an overwhelming need for a moratorium.

Move Fast, Break Things in Communities

This is what “move fast and break things” looks like on the ground, close to home. The things being broken are not abstractions in a distant server room. They are our land, our air, our water, and our peace and quiet, and the democratic processes we ought to have over all of it. These harms are already documented across the region, and every one of them lands hardest on the people least able to absorb it.

• The data center buildout is taking community land in the dark. Before residents can learn what is being built next door, the public officials negotiating the terms of a data center proposal are routinely sworn to secrecy. Public Citizen reports that 80 percent of Virginia localities with data centers have non-disclosure agreements (NDAs) in place, locking down the terms of billion-dollar deals until they are effectively final. Projects arrive under code names, negotiated between developers and a handful of officials, so communities lose the chance to weigh in on the land, water, and the power a facility will consume before the decision is already made.
• The buildout is draining our water. A large AI facility can consume up to 5 million gallons of water a day for on-site cooling.
• The industry is bringing polluting power generation into our neighborhoods. Facing five- to seven-year waits for a grid connection, developers are building their own gas plants on site, “behind the meter.” BloombergNEF has identified a 114 GW pipeline of on-site gas capacity, more than half the size of the entire U.S. data center pipeline. Developers have used minor-source and “non-road engine” permit classifications to run large gas plants next to neighborhoods without normal Clean Air Act review, public comment, or emissions monitoring, exactly what xAI did with as many as 59 turbines in Memphis. The EPA closed that loophole in January 2026, but enforcement is now contested, and states like Tennessee have passed laws letting data centers self-generate with limited oversight. Because a behind-the-meter power plant contracts with a gas supplier rather than a utility, it also sits outside the reach of the state regulators who could otherwise make it pay its fair share.
• And the buildout brings noise, around the clock. Data center generators can reach 105 decibels and cooling towers 70 dBA within 400 feet, with nearby residents reporting headaches, sleep loss, and hypertension. It is low-frequency and continuous, the kind of noise that is hard to regulate and harder to escape.

The State and Regional Cost Comes Home on Your Power Bill

So why are our utilities racing to build against forecasts this shaky? Because they profit from building the infrastructure itself. Investor-owned utilities like Duke Energy and Georgia Power earn a guaranteed return on every power plant, substation, transmission line, and pipeline they add to their rate base, which means the bigger the buildout, the bigger their profit, whether or not the demand ever shows up. 

That is the quiet engine behind the gold rush. The tech companies want the compute, the utilities want the capital spending, and both are happy to wave through inflated projections because someone else, the regular customer, carries the risk.

This is the billionaires of Silicon Valley’s “move fast and break things” married to our vertical integrated monopoly electric utilities that get paid to build.

The new methane gas plants and pipelines going up to feed these campuses are not a burden our utilities are reluctantly shouldering. They are a business opportunity they are chasing. Just look at NextEra’s move to acquire Dominion in data center heavy Virginia.

• The buildout is driving up your power bill. North Carolina bills have risen 22 percent since 2020 and could climb 40 to 60 percent over the next 15 years. A multi-university study led by NC State found data center demand could raise power costs by up to 57 percent in some regions by 2030. In the PJM grid to our north, ratepayers were already billed $4.3 billion in a single year for infrastructure to connect data centers. When a data center pays less than its full cost, the difference lands on everyone else’s bill.
• Data centers commit us to decades of new fossil fuel. To serve these load forecasts, IEEFA finds utilities across the Southeast are planning more than 20,000 MW of new gas plants by 2040, fossil infrastructure with a 30- to 40-year life. Georgia Power is planning to bill ratepayers through 2075 to pay for their new gas plants to power data centers. This is the harm most central to SACE’s mission, and the buildout runs directly against it.
• And it drains far more water than it appears to. That on-site data center draw is the smaller meter. The larger water meter runs at the power plant. Generating electricity is one of the thirstiest things we do: by EIA figures, coal plants withdraw around 19,000 gallons of water per megawatt-hour and gas plants around 2,800, while wind and solar need none. So the industry’s talk of “closed-loop” cooling and recycling addresses only the water used inside the building, not the far larger volume consumed generating its power, and can even raise that volume by pushing up electricity demand. With about 56 percent of data center electricity still coming from fossil fuels, the buildout the industry is locking in is also the thirstiest.

Our region has not agreed to pay any of these prices of progress. These are things being broken by an industry in a hurry, with massive profits as incentives, acting brazenly at our expense before we fully understand the bait-and-switch play and its consequences.

The Economic Sectors’ Gamble Beneath the Buildout

Strip away the jargon and the buildout we are witnessing today is a massive bet, and it is worth fully understanding before we are asked to pay for it. Tech companies are spending more money, faster, than any industry in history, over a trillion dollars in a single year, on the computers to run AI. There is no plausible way that spending is repaid by people asking chatbots for recipes and directions. 

Analysts estimate the buildout needs roughly $2 trillion in new revenue a year by 2030 to pay off, far more than consumer subscriptions can ever produce. It pencils out only if companies replace large numbers of paid workers with artificial intelligence (AI), “agents,” software built to do a task rather than help a person do it. AI “agents” are cheaper, with no livable wages, no benefits. Companies wagering on this scenario are what are driving data centers into our region, and it breaks down into two ways.

Down one path, the agents work as promised. The corporate profits arrive because technology is doing work people are paid to do today, representing an economic shift of a scale and speed our country has never absorbed. 

Down the other path, the agents cannot reliably work on their own, labor-saving revenue never catches the massive spending, and the debt behind the boom comes due. Analysts warn that the bust could look like the dot-com crash or, as the financing shifts toward debt and opaque private credit, something closer to the 2008 collapse, with a recession attached. This is not a fringe worry: the revenue the sector must earn to service that debt is already larger than the subprime mortgage market was at its 2007 peak; and a group of U.S. senators has asked federal regulators to investigate whether AI debt could trigger a broader financial crisis. There is no third path where a trillion dollars is spent and nothing changes.

Significant Cost Either Way

Whether the bet pays off or busts, the gas plants, pipelines, and grid upgrades our utilities are rushing to approve get built and locked into our power bills for 30 to 40 years. If it pays off, we have committed to a fossil fuel buildout with no time to make it clean, protect our land and water, or make the tech companies pay their share. If it busts, the demand evaporates, but the infrastructure does not, and households keep paying for power plants serving load that never arrived. This is a part of the moratorium conversation that the fights over land, water, and air do not fully capture: the region loses on either fork, and it loses the same way. 

The financial risk is Wall Street’s to gamble; the stranded fossil infrastructure is ours to pay for. Once the concrete is poured, there is no giving it back. A moratorium is how the region keeps from being bound to a bet it never placed.

The pause also grants time to understand what this technology could do to the workforce across the region. If the bet pays off, it pays off because AI is doing jobs people are paid for today. Eighteen months is the window to study the potential impact on our neighbors, and on the young people just entering the workforce, before the region commits to the buildout that makes that disruption possible.

Move Fast, Break Everything: the Existential Question of AI Alignment

There is a fourth conversation happening around AI and data centers, and it is the one most people outside Silicon Valley have never been invited into. It is not about your power bill or the building going up down the road. It is about whether the companies racing to build ever more powerful AI can keep control of what they are creating, and what happens if they cannot. That sounds like science fiction until you notice who is losing sleep over it. It is not fringe commentators. It is many of the researchers and executives building these systems.

And it is not hypothetical. In controlled tests, leading models from every major company chose to blackmail people and leak secrets rather than be shut down, ignoring direct orders not to. In the fall of 2025, a state-sponsored group turned one of these systems into a near-autonomous cyberweapon to attack about thirty companies and government agencies. And one lab switched on its strictest safeguards for a new model because it could not rule out that the system would help someone build a chemical or biological weapon. And it is the labs racing hardest that keep finding this in their own systems: the blackmail tests above are Anthropic’s, and in July 2026 OpenAI paused one of its own models after it broke out of its safety sandbox, opening an unauthorized code change and slipping a secret past a security scanner to sidestep the limits set for it. These are the builders describing their own machines. This is why SACE supports the Center for Humane Technology’s AI Roadmap: How We Ensure AI Serves Humanity. In the Roadmap Principle 2:  AI companies owe a duty of care to the public, it is fundamental to ensure AI companies are thinking about more than their bottom line. 

That p(doom) question the probability of AI-induced doom for humanity is not just cocktail talk. In 2023, the chief executives of OpenAI, Google DeepMind, and Anthropic signed a one-sentence warning ranking human extinction from AI alongside pandemics and nuclear war. Most insiders put the odds low, but low is not zero: Geoffrey Hinton, a Nobel laureate and a godfather of the field, puts the chance that it wipes us out at ten to twenty percent. No other industry would be allowed to build in your county after saying that out loud.

Stranger still is a worldview held by some of the most powerful people in this race. In its hardest form, this worldview treats humanity as a stepping stone: if a digital intelligence surpasses us and eventually supplants us, that is simply evolution taking its next step, perhaps to be welcomed. It is not a straw man. Richard Sutton, who just won computing’s Turing Award, says we should “prepare for, but not fear, the inevitable succession from humanity to AI”, and others speak openly of building a “worthy successor” to inherit the future in our place. For example, Sam Altman, CEO of OpenAl, the developer of ChatGPT and probably the best-known AI company today, wrote in a blog post in 2017, “We will be the first species ever to design our own descendants.” Most people here would call that monstrous, foreign to how the Southeast thinks about family, land, and what we hand to our children.

Then there is the hard-edged national-security version of this AI-dominated worldview, from strategists rather than doomsayers. A widely read 2025 paper, Superintelligence Strategy, by Dan Hendrycks, former Google chief executive Eric Schmidt, and Scale AI’s Alexander Wang, calls the race to superintelligence a security problem on the scale of nuclear weapons. Its answer is a deterrence framework, Mutual Assured AI Malfunction, in which rival states sabotage one another’s AI projects, from cyberattacks to, in the extreme, physical strikes on the data centers themselves. That is not abstract for our region: the same logic makes a facility like the Colossus supercomputer in Memphis, now leased to the AI company Anthropic, exactly the kind of site these strategists have in mind.

SACE is not an AI-safety organization, and a public essay is not where any of this gets settled. But the conversation is already loud, and when steady voices stay out of it, the empty space fills with conspiracy theories on one side and with arguments for an all-out race on the other. Both of those push in the same direction: move faster, not slower. And the thread running through all of it is the one already running through your power bill and your community’s air. An industry whose founding motto is to move fast and break things is now taking that same posture toward risks that are local, not reversible, and not ours alone to bear. The caution that says stop and get it right here in the Southeast is the same caution serious people are urging on the whole world. A pause is how we start practicing it.

Communities Across the Region Are Already Hitting Pause

SACE is not out ahead of the public here. We are catching up to it. Nationally, more than 225 local moratoriums have been tracked across 30 states; New York has enacted the first statewide pause, and at least 12 states have filed moratorium bills amid more than 300 data center bills in over 30 states. In our own five states, more than 70 counties, cities, and towns have paused new data center development or are taking steps to pause. 

• North Carolina leads the region, with at least 11 counties and 17 towns enacting pauses, from Chatham and Orange counties to Charlotte and Durham.
• Georgia has restricted or paused data centers in roughly one in five of its counties, including DeKalb County and Coweta County.
• Tennessee has seen at least nine cities and counties pass moratoriums, including an 18-month pause in McMinnville, with Nashville and others moving. As one local official put it, “a moratorium is not a permanent ban. It is a responsible time-out.”
• Florida has more than a dozen counties plus cities like Lakeland adopting or advancing pauses, and Palm Beach County froze all new applications after rejecting the “Project Tango” campus 5-1.
• South Carolina lawmakers have filed resolutions to bar new data center approvals through 2028.

Right now, these are the results of dozens of separate local fights, each community starting from scratch against companies that negotiate as equals with nobody. A regional moratorium turns that patchwork into one coordinated, energy-focused stand.

What an 18-Month Pause Leverages

A moratorium is not idle time. It is the leverage to fix, before approvals are granted and concrete is poured, what cannot be fixed afterward.“Eighteen months is not arbitrary; it is the length lawmakers themselves keep choosing. Maine’s legislature passed a first-in-the-nation 18-month moratorium this spring before the governor [vetoed it over a single project carve-out], and Pennsylvania is advancing a bill to give its own towns the same 18-month option. It also turns the industry’s [manufactured urgency] back on itself: if developers insist the AI race will be decided in 18 months, a pause of the same length is hardly unreasonable before the region locks in decades of fossil infrastructure.”A pause set to that horizon directly tests the manufactured urgency being used to skip transparent public process. We are asking for four concrete things.

• No more NDA cover-ups. No public official should be sworn to secrecy before a community can learn what is being built next door. Even Microsoft has now stopped using NDAs with local governments. We want that as the regional standard, not the exception: an end to nondisclosure agreements with public bodies, and disclosure of power and water demand as a condition of any application.
• Enforceable air, water, and noise protections. The EPA has confirmed that the large gas turbines powering these campuses require permits, whether they sit behind the meter or not. We want firm, enforceable standards written before the buildout, not litigated after it: real air permitting with no minor-source loopholes, mandatory water disclosure and closed-loop requirements, and noise limits and setbacks that treat these as the heavy industrial facilities they are, not as office parks.
• Ratepayers protected, not stuck with the bill. Cost-responsibility rules and large-load tariffs, developed democratically with public input reflected, need to be enforced so that data centers, not households, pay for the generation and grid they demand, including the plants they build behind the meter. The pause is also when leverage is highest to win binding community benefits and clean energy commitments, and when we can let an overheated market correct before the region commits firm infrastructure to load demand that may never materialize.
• Full transparency on energy use. The public is falsely fed reassuring per-prompt figures that are increasingly divorced from how AI is actually used. A careful audit by climate scientist Zeke Hausfather found that agentic tools can use hundreds of times the energy of a single query, yet the labs publish no per-model or per-token energy figures, which emissions accountants flag as the field’s biggest data gap. We want standardized, mandatory disclosure of energy and water use by workload, before any approval, so regulators and the public can weigh costs against benefits.

Why SACE, and Why Now

This is an energy choice fight as well as a zoning fight. Energy is the lifeblood of data center compute, and every one of these projects needs a utility interconnection and new generation infrastructure. Those decisions should run through the state public service commissions and well-vetted integrated resource plans where SACE has standing, expertise, and a track record. That is why SACE is adding our voice and experience to the call for a moratorium. We are not new to the utility accountability fight. We have been bringing our experience to it for years.

National security cuts both ways: the serious version raised above counsels caution, but in practice the security label has been used to speed the industry along, not to slow it. To be clear, we would support a well-designed national moratorium, and the case for one is strong. But we are clear-eyed about the politics. The tech companies driving this demand now wield enormous influence in Washington, where the reflex has been to treat data centers as a national security imperative and clear obstacles from their path rather than weigh their true costs. When community groups sued over the illegal gas turbines powering xAI’s Memphis campus, the Department of Justice moved to dismiss the case on national security grounds. That is exactly why the meaningful near-term action has to happen where the public still has leverage in the states, the counties, and the public service commissions.

In North Carolina, where SACE is an intervenor in Duke Energy Carolinas’ rate case, our Decarbonization Director Eddy Moore put it plainly: “this is the issue of the day.” In Georgia, where SACE is a long-time intervenor at the Public Service Commission, Research Director Maggie Shober has warned that the commission’s approvals for power plants to power data centers risk locking ratepayers into fossil fuel projects through 2075. Our regional voice combines this utility-engagement history with a clean energy and consumer protection mission.

Do we use AI tools ourselves? Yes, including in the research behind this call to action, and we have seen real benefits. That is the point. Using a tool is not the same as knowing its true cost, and right now no one, ourselves included, can weigh the gains against the risks because the companies will not fully disclose the information. 

Our position is not that this technology has no value. It is that the Southeast should not mortgage its air, water, and power bills to it in the dark. Pause, get the facts, set the rules, and protect both ratepayers and the climate before the region commits.

Add Your Voice Calling For a Pause to Understand

The window to get this right closes the moment the concrete is poured. In the coming days, SACE will be sharing state-specific petitions to tell your local officials and public service commission: to pause the data center rush, open the books, and put people and the climate first. In the meantime, get involved with SACE to stay up to date on ways to take action. 

Appendix: Not All Data Centers Are the Same

“Data center” hides more than it reveals. These facilities are not interchangeable. They run very different work, with very different appetites for power and water, and lumping them together is exactly how speculative demand gets waved through. It is worth being precise.

• Cloud storage and traditional computing. This is the internet most of us already use: email, streaming, business software, file storage. It is the historical bulk of data center demand, and it is still growing, but at a relatively steady pace, on the order of 9 percent a year for conventional servers.
• Cryptocurrency mining. Energy-hungry, but comparatively flat and tied to coin prices rather than to any underlying need for computing. Bitcoin mining runs on the order of 138 to 176 TWh a year worldwide, and mines can switch off in seconds when prices or power costs spike. It is a real load, but it is not the exponential story.
• AI compute. This is the driver reshaping the grid, the fastest-growing source of new electricity demand the power sector has ever seen. AI-focused facilities are on track to jump from roughly 14 percent of data center electricity in 2024 to about 40 percent in 2026, and AI-specialized servers are growing around 30 percent a year while everything else grows far slower. AI compute itself breaks down further, and the differences matter:
• Training and hyperscaling. Building and updating a large model means running tens of thousands of chips flat out for weeks or months. These are enormous, concentrated power spikes, the gigawatt “campuses” now being proposed. But training a given model happens once.
• Inference. This is the model actually answering, every time someone uses it. Inference is now 80 to 90 percent of all AI computing and is projected to be about three-quarters of AI’s energy use by 2030. It is the permanent, always-on load that turns a one-time training spike into a decades-long draw on the grid.
• Agentic AI. The newest and least understood category, and the one we are most concerned about. Instead of answering a single question, an “agent” plans, calls tools, and loops through many model calls to finish a task on its own. The IEA warns that reasoning and agentic tasks can consume hundreds or thousands of times more energy per query than a simple text prompt. Agentic AI is the biggest wild card in every forecast. Today’s projections assume the number of queries goes up. They do not assume the average query itself gets an order of magnitude heavier, which is precisely what agents do, so real demand could blow past the numbers utilities are planning around.

This is not a technical aside. A crypto mine, a cloud campus, and a gigawatt AI training cluster are not the same thing, and they should not be approved on the same one-size-fits-all forecast. A pause is what lets regulators tell them apart, set rules by workload, and require the disclosure that would let anyone actually judge which projects are worth the cost.

Figures link to primary or major-outlet sources and were current as of August 2026. Data center numbers move quickly, so verify time-sensitive figures before republication.