A 9-Gigawatt Datacenter Just Got Approved. Let That Sink In.

In May 2026, commissioners in Box Elder County, Utah approved a project called Stratos: a 40,000-acre AI datacenter complex backed by investor Kevin O'Leary that will generate and consume 9 gigawatts of power. For context, Utah's entire current electricity grid runs at roughly 4 gigawatts. This single facility will produce more than twice the power the entire state currently uses, drawing it from the Ruby Pipeline, a 680-mile interstate natural gas line running from Wyoming to Oregon.

This is not a proposal. It is approved. Construction is expected to begin soon, with the first gigawatt coming online within two years.

What makes Stratos notable is not that it is uniquely extreme. It is notable because it represents the direction the industry is moving: at scale, fast, and powered by whatever fuel is available and cheap.

Energy Gets the Headlines. Water Is the Bigger Problem.

When people talk about AI's environmental footprint, they usually talk about electricity. That conversation matters. Global datacenters consumed 415 terawatt-hours of electricity in 2024, and that number is projected to double by 2030. AI-specific systems alone could generate between 32.6 and 79.7 million tons of CO2 in 2025, a figure comparable to New York City's entire annual carbon footprint.

But the water story is less visible and, depending on where you live, potentially more urgent.

Large AI-focused datacenters can consume up to 5 million gallons of water per day for cooling. Of that water, approximately 80% is evaporated and never returned to the local watershed. Global datacenter water consumption is projected to grow from 560 billion liters annually to 1.2 trillion (yes, with a T) liters by 2030. Some researchers project that water use for AI cooling could increase by as much as 870% under current trajectories.

What happens when companies draw that water without meaningful accountability? In a county south of Atlanta, a Blackstone-owned datacenter called QTS drew approximately 29 million gallons of water over 15 months through two unauthorized connections the county did not know existed. Residents first noticed when water pressure dropped in their neighborhood. When the investigation concluded, the company owed $147,474 in unpaid charges. There was no fine.

That is the enforcement environment these facilities currently operate in.

This Is Not a Utah Problem. It Is a North Carolina Problem.

The Appalachian Basin is being actively marketed to the AI industry as the next major datacenter hub. The selling points are straightforward: cheap land, established natural gas infrastructure, cooler mountain air for lower cooling costs, and regulatory environments that have not yet built up meaningful resistance to industrial-scale projects.

In January 2026, county commissioners in Stokes County, NC rezoned 1,845 acres of residential and agricultural land for heavy industrial use to accommodate a hyperscale datacenter complex called Project Delta. Community groups, the Dan River Basin Association, the National Hairston Clan, and the Southern Environmental Law Center filed suit arguing the rezoning lacked procedural grounding and ignored documented concerns about water, air quality, noise, and sacred sites tied to Saura tribal history. The case is ongoing.

Amazon has broken ground on a $10 billion campus in Richmond County, NC: 1,200 acres, 20 buildings. Microsoft acquired 1,385 acres in Person County. These are not edge cases. They are the early pattern of a much larger buildout.

The dominant energy source for this expansion is natural gas. That means new pipeline infrastructure and a functional lock-in of fossil fuel generation at exactly the moment the region had begun transitioning away from coal. One energy researcher described the datacenter rush as making it "functionally impossible" to meet zero-carbon electricity goals on the Appalachian grid. Dominion Energy has already announced plans to bring 5.9 gigawatts of additional natural gas generation online between 2030 and 2036 specifically to serve datacenter demand.

The Appalachian region bore the environmental cost of a century of coal extraction. It is now being positioned as the resource base for the next extractive cycle.

The French Broad Does Not Have 5 Million Gallons a Day to Spare

Western North Carolina draws more than 95% of its water from surface sources. The French Broad, Watauga, Little Tennessee, Hiwassee, and New rivers all originate in the Southern Blue Ridge. These are not isolated waterways. They supply municipalities, farms, and households across the region and feed into larger river systems downstream.

The fractured bedrock aquifer system underlying the mountain terrain is not a deep reservoir sitting in reserve. Average well yields in the mountains run around 23 gallons per minute, a figure that reflects the geology of the region, not a policy choice. There is no hidden buffer.

A large datacenter drawing 5 million gallons per day would consume the annual water equivalent of approximately 50,000 mountain households. Cooling discharge does not simply disappear, either. It can carry heat, heavy metals, and chemical treatment compounds into whatever waterway receives it, degrading river health in ways that affect aquatic ecosystems and downstream water quality for years.

The Blue Ridge has no margin here.

We Already Know What Happens When This Region Gets Hit

In September 2024, Hurricane Helene made landfall and moved inland. When it reached the mountains, it produced three-day rainfall totals with less than a 0.1% probability of occurring in any given year. The French Broad corridor flooded catastrophically. Roads became creek beds. Bridges failed. The Asheville water system went down.

108 people died in western North Carolina. More than 125,000 housing units were damaged or destroyed. Total damage reached $59.6 billion. The Blue Ridge Parkway closed. Recovery is still ongoing more than a year later.

Before Helene, Asheville was described in national media as a climate haven: a place people were moving to escape sea level rise, coastal storms, and extreme summer heat in the South. Climate researchers now describe Helene as a "gray swan" event, a situation where a place not previously considered high-risk turned out to be anything but immune from extreme weather made more severe by a warming climate.

The technology angle on this is worth sitting with. Spruce Pine, NC sits in the Blue Ridge and produces 70 to 90% of the world's high-purity quartz used in semiconductor manufacturing. The quartz from Spruce Pine goes into the process that produces the silicon crystals inside the chips that power AI systems. Helene damaged those operations and temporarily threatened the global semiconductor supply chain. The same region now being eyed for AI datacenter infrastructure nearly lost the mines that supply the raw material for the chips those datacenters run on.

This is not an abstraction. We know exactly what climate disruption looks like here. We watched it in 2024, and the communities in the mountains are still rebuilding.

The Math Does Not Work Out Well

A Cornell University study published in late 2025 identified a roadmap that could cut AI datacenter carbon emissions by 73% and water consumption by 86% compared to worst-case projections. The path involves smarter geographic siting, faster grid decarbonization, and operational efficiency improvements. The researchers specifically identified the Midwest and wind-belt states as better environmental fits for large-scale AI infrastructure.

The research is credible. The roadmap is technically achievable. The current trajectory does not follow it.

The pattern right now is: find land that is cheap, find power that is available, move fast, and let communities and regulators catch up later. Stokes County had to file a lawsuit to force a basic procedural review. The Georgia water case resulted in a bill, not a fine. These outcomes are not anomalies. They reflect how lightly regulated this buildout currently is.

The communities that push back get outgunned by capital. The ones that do not push back live with the infrastructure for decades.

We Are Not Anti-AI. We Are Pro-Accountability.

This piece is written by someone whose livelihood depends on people using AI well. We help businesses automate workflows, build AI-assisted tools, and make smarter decisions using AI every week. We are not arguing against AI. We are arguing against using AI without looking at what it costs, physically, to run.

Here is what responsible AI use looks like from where we sit:

Ask your AI vendors about their energy and water sourcing. Real renewable energy commitments are verifiable. Carbon offsets purchased to balance natural gas consumption are not the same thing, and you should know the difference before you treat a vendor's sustainability claims as meaningful.

Use AI where it actually replaces high-volume, repetitive work. Do not reach for AI as the default response to every business problem. That is how individual decisions aggregate into infrastructure demand that burns through water and gas to produce outputs that could have been handled more efficiently. Every query has a cost. Not every query is worth it.

Build an AI policy for your organization. A good AI policy does not just address accuracy, hallucination risk, and data security. It addresses proportionality: what are we using AI for, and is the cost of running it proportional to the value it creates? We have written about what a practical AI policy looks like in our AI policy article, and it is worth reading if you have not.

The Blue Ridge is not a datacenter hub waiting to happen. It is a watershed. It is a forest. It is the place that people who work here and live here have spent the past year rebuilding after one of the worst natural disasters in the region's recorded history.

What happens to it next depends in part on the decisions that businesses, policymakers, and the AI industry make in the next few years. Those decisions should be made with eyes open about what the current trajectory actually looks like.

Sources and Further Reading