How AI Affects the Environment: Water, Labor, and Local...

The week's environment and sustainability reporting outside the interconnection fight. Data center water use is on track to double or quadruple by 2028, a skilled electrical labor shortage is capping the build-out, New York has a deliverables list to finish during its permitting pause, and Amazon published facility-level power and water numbers while most developers are still negotiating under nondisclosure agreements.

Our feature this week follows the bargain taking shape at the interconnection queue, where a data center can now buy grid access with heat pumps, capacity contracts, and hardware that rides out a fault. The rest of the week's environment and sustainability reporting landed a layer away from that queue: on water, on a labor shortage with no quick fix, on what states are writing down while they wait, and on what communities are being asked to accept without reading the paperwork.

Water Is Turning Into the Second Interconnection Constraint

Utility Dive covered a Kansas Health Institute report that puts direct US data center water consumption at 17 billion gallons in 2023, a figure the authors say could double or quadruple by 2028. Twenty percent of US data centers already sit in regions the report classifies as water-stressed.

Texas shows the curve at state scale. Data centers there used roughly 49 billion gallons last year and could reach 399 billion by 2030, according to figures The Next Web compiled. That is an eightfold move inside a decade, in a state that has spent the past several summers rationing water for other reasons.

The report also breaks down what is currently powering the load: about 56 percent fossil, 22 percent renewables, 21 percent nuclear. Any honest accounting of the environmental effects of artificial intelligence has to carry that mix alongside the water, because the two costs move together. Evaporative cooling trades water for electricity, and air cooling trades the other way.

Water has no equivalent of the mechanism the feature article covers. There is no aggregator bundling residential water savings into a verified credit a developer can buy, and no queue position to win with one. Siting a large load into a stressed basin is a specific decision made by specific companies, and it is the decision most likely to end up in front of a county commission.

For teams that want basin-level and grid-level data inside a siting or reporting workflow rather than in a PDF, AgentPMT's Climate, Environment and Land Data Hub and Global Energy and Power Grid Data catalog entries are the two connectors built for exactly that pull.

Source: Utility Dive

The Ceiling on the Build-Out Is Skilled Electrical Labor

BloombergNEF's revised 2035 forecast got the headlines this week, and the feature article deals with what that number does to capacity planning. Buried in Latitude Media's coverage was a constraint that no forecast revision fixes.

Mark Daly, BNEF's head of technology, put the bottleneck in trade terms: "You have to be able to hire people who know how to do really, really, really sophisticated electrical work." Not chips, not turbines, not transformers on a lead time. Electricians who can commission medium-voltage switchgear without burning down a substation.

BNEF also tracks 124 GW of announced on-site gas capacity. Announced is a press release. Someone has to pour the pads, pull the cable, and sign off on the protection scheme, and that same trade is simultaneously being asked to install residential heat pumps under New Jersey's new law, build out the battery capacity a PJM contract would finance, and wire up several gigawatts of uninterruptible power systems.

Of every constraint in this week's reporting, this is the most tractable. Where energy and artificial intelligence meet, the shortage is a training pipeline with ten years of visible demand attached to it, which is a far easier thing to fund than a new transmission corridor. Utilities, community colleges, and the developers competing for the same crews all have an obvious reason to pay for it.

Source: Latitude Media

New York's Pause Comes With a Homework List

The feature covers why New York reached for a moratorium. What the state actually has to produce during it is a separate and more useful document.

The pause runs one year and applies to state environmental permits for hyperscale facilities capable of drawing 50 MW or more. Inside that year, New York has committed to producing a Generic Environmental Impact Statement covering energy demand, water use and water quality, and air quality for the facility class as a whole.

Alongside it comes a Community Investment Framework meant to guide local negotiations. Its named scope includes infrastructure, childcare investment, and prevailing wage standards. Childcare in a data center siting framework is genuinely new, and it signals that the state read the local objections as economic rather than purely environmental.

State Senator Kristen Gonzalez framed the year as time to plan rather than time to stall. The distinction is testable: a Generic Environmental Impact Statement is a reusable artifact. Once it exists, a developer in New York negotiates against a published baseline instead of relitigating energy and water questions from zero at every county hearing, which is the part of the current process that burns quarters.

Other states weighing similar bills will copy whichever document ships first. That makes the quality of New York's GEIS a national question, not a New York one.

Source: New York Governor's Office

Bills, Wells, and Nondisclosure Agreements

The July 18 national day of action drew the week's coverage. The material underneath it, reported by TIME and The Next Web, is more specific and more fixable than "people are angry."

Start with bills. In PJM territory, annual electricity supply costs jumped from 2.2 billion to 14.7 billion dollars in a single auction. Translated to a household, that reads as an estimated 21 dollars a month more in Washington DC, 18 in western Maryland, 16 in Ohio. You do not need a briefing on transmission economics to notice a line item like that on your own bill.

Then wells. TIME reported from Coweta County, Georgia, where residents Laura Beth and Steve Swope live near a proposed site. Steve Swope on what he expects: "Our wells will either run dry or they'll become polluted and caked with sediment and mud." Whether or not the hydrology bears that out, a household on well water has no utility to complain to and no meter to point at.

Then the paperwork. Roughly 80 percent of Virginia localities use nondisclosure agreements on data center proposals. That is a choice made by developers and accepted by counties, and it is the single most self-defeating practice in this entire category. A resident who cannot read the proposal will assume the worst thing in it, and they are usually right that something was worth hiding.

The cost of that posture is now legible. More than 300 cities, towns, and counties have enacted bans or moratoriums, and 75 major projects worth over 130 billion dollars were delayed or canceled in the first quarter of 2026. About one-third of Americans approve of the current pace of construction, and 14 percent say they would be comfortable with a facility near them. Publishing the proposal is cheaper than losing the site.

Source: TIME and The Next Web

Developers Report That Power Anxiety Peaked

This one comes with a label attached: it is a Latitude Studios survey produced as partner content sponsored by Bloom Energy, a company that sells on-site power generation. Read it as direction, not measurement.

With that caveat, the shape is interesting. Across 156 respondents spanning hyperscalers, colocation providers, neoclouds, developers, and chip companies, 51 percent now rank power availability among their top three site-selection factors. In June 2025 that figure was 84 percent.

A drop that size in thirteen months does not mean power got easy. It more likely means the respondents already spent the money. Thirty-three percent report investing in grid support and resiliency, and 29 percent in minimizing water consumption, which is roughly what you would expect from a cohort that solved its own interconnection anxiety with capital.

Twenty-eight percent now cite worsening local opposition, and respondents collectively expect one-third of US data centers to be running entirely on on-site power by 2030. Given who paid for the survey, treat that last projection as a sales forecast wearing a research jacket. The 84-to-51 move is the part corroborated by behavior elsewhere in this roundup.

Source: Latitude Media (Latitude Studios, sponsored by Bloom Energy)

Amazon Published Its Numbers Instead of Waiting to Be Asked

Latitude Media ran a conversation with Brandon Oyer, AWS director of energy and water strategy, that functions as the counter-example to the NDA posture above. Amazon disclosed specifics, and specifics are checkable.

The portfolio figures: more than 700 renewable projects globally totaling 42 GW. Siting decisions now prioritize grid locations where the added load most benefits existing infrastructure, which Oyer credits for investments in Jackson, Mississippi and South Bend, Indiana. That inverts the usual siting logic, which optimizes for cheap land and available fiber and treats grid impact as someone else's arithmetic.

On water, Amazon reports water use effectiveness improved 52 percent between 2021 and 2025, putting it roughly 75 percent of the way to a 2030 water-positive goal. Project Rainier, a 2.2 GW campus in Indiana, runs water cooling only two to three percent of the year and uses outside air the rest. Beyond the fence line, 235 million dollars went to Oregon water infrastructure, and precision irrigation support cut Mississippi farm withdrawals by about 150 million gallons a year.

On generation, a 500 million dollar investment in X-energy targets small modular reactors in 80 MW increments, and Amazon signed the White House Ratepayer Protection Pledge committing to cover its own electricity production costs rather than socializing them.

Whether every number survives audit is a fair question for the next reporter. What matters for anyone else building at this scale is that the numbers exist and are published at facility granularity. Sustainable AI development turns out to be a measurement discipline before it is an engineering one, and the operators who can already answer what a site draws, how much water it returns, and what it contributes back are the ones positioned to take the deals the feature article describes. The ones still negotiating under an NDA will be explaining themselves to a county commission instead.

Source: Latitude Media

Sources

US data centers could 4x water use by 2028, double electricity by 2030, Utility Dive
BNEF nearly doubled its forecast for US data center power demand, Latitude Media
Governor Hochul Highlights First Statewide Moratorium on New Hyperscale Data Centers, New York Governor's Office
In Western New York, Governor Hochul Highlights First Statewide Moratorium on New Hyperscale Data Centers, New York Governor's Office
Community Backlash to AI Data Centers Is Growing Across the U.S., TIME
Data centre backlash goes national as protests hit 42 US states, The Next Web
Survey: As community scrutiny grows, data center developers turn to on-site power, Latitude Media (Latitude Studios, sponsored by Bloom Energy)
Inside Amazon's approach to data center sustainability: A conversation with Brandon Oyer, Latitude Media