WHAT TO DO ABOUT DATA CENTERS.
I asked Chatgpt how to solve the problems created by so many new Data Centers, and this is what it said.
Deseradi, 8/24/26
There is a lot of anger right now against data centers because they use so much electricity and water that sometimes the cost of electricity goes up and the water becomes bad for humans. How real is this problem in the US? And how can it be solved?
Yes. The problem is real, but the strongest version of the claim—“data centers are making electricity and drinking water unaffordable everywhere”—is too broad. The impacts are highly local. In some communities, especially places with constrained electric grids or scarce water, the effects can be substantial. In others, a new data center may have relatively little effect on household bills or water supplies.
What has changed recently is the scale and speed of construction, particularly for AI. The numbers are large enough that this is no longer just a theoretical environmental concern.
1. Electricity: this is probably the more immediate national problem
Lawrence Berkeley National Laboratory's 2025 update estimates that U.S. data centers could consume 11.8% of all U.S. electricity by 2030, with a plausible range of 9.5%–15.3%. That is an enormous increase from about 4.4% in 2023.
And the problem isn't simply the amount of electricity. It's where and when the electricity is needed.
A giant AI data center can suddenly add hundreds of megawatts of demand to a local grid. If the grid doesn't have enough generation or transmission capacity, somebody has to pay for:
new power plants
transmission lines
substations
distribution infrastructure
grid upgrades
additional reserve capacity
This is already happening. The EIA says data centers are driving much of the current growth in U.S. electricity demand, with particularly rapid growth in Texas and the PJM region. Its modeling found that unusually rapid demand growth could push Texas's 2027 wholesale electricity price 79% above its baseline forecast.
PJM, which serves much of the Mid-Atlantic and Midwest, is also facing rapidly increasing demand and has projected roughly 70 GW of additional demand by 2038, much of it from large new energy users such as data centers.
So when residents say:
"Why should my electric bill go up because Amazon/Google/Microsoft wants to build an AI factory?"
That is a legitimate policy question.
And there is an important distinction here: a data center doesn't inherently have to raise residential rates. It raises rates when the utility's rate structure or regulatory decisions allow some of the infrastructure costs to be shifted onto ordinary customers.
2. Water: also real, but much more geographically dependent
The water issue is somewhat different.
Data centers use water primarily because computers generate enormous amounts of heat. Traditional evaporative cooling can consume substantial quantities of water.
A recent USGS synthesis notes that a 100-MW data center can consume around 2 million gallons of water per day, although actual consumption varies enormously with the cooling system and climate.
And some of that water is genuinely consumed, rather than merely withdrawn and returned. Water evaporated during cooling doesn't go back into the local water system.
The problem becomes particularly serious when a data center is put somewhere like:
Arizona
Nevada
parts of Texas
Southern California
other drought-prone regions
rather than somewhere with abundant water.
A 2026 study found that data-center cooling could require an additional 697 million to 1.45 billion gallons of peak water capacity per day nationwide by 2030 under some scenarios. Importantly, the researchers emphasize that the burden would be highly concentrated in the communities hosting the facilities.
That's a crucial point.
The national average can be misleading.
Suppose data centers consume a relatively modest percentage of America's total water.
That doesn't mean much to a town whose water system is already operating near capacity and suddenly has a customer wanting hundreds of thousands or millions of gallons a day.
It's analogous to electricity:
Nationally: perhaps manageable.
Locally: potentially enormous.
3. But there is an important misconception about "ruining the water"
I'm more cautious about the claim that data centers generally make water "bad for humans."
That's not normally what happens.
The bigger concern is depletion and competition for water, rather than a data center routinely contaminating a community's drinking-water supply.
Cooling systems can produce wastewater containing concentrated minerals and treatment chemicals, and poorly managed discharge can create environmental problems.
But the much more common concern is:
"Is this facility consuming water that our community cannot afford to lose?"
That's a very different problem from:
"Is this facility poisoning our drinking water?"
The first is a significant and growing issue in some locations. The second should not be generalized to data centers as a whole.
4. The really interesting thing is that many of these problems are solvable
Lawrence Berkeley National Laboratory's 2025 update estimates that U.S. data centers could consume 11.8% of all U.S. electricity by 2030, with a plausible range of 9.5%–15.3%. That is an enormous increase from about 4.4% in 2023.
And the problem isn't simply the amount of electricity. It's where and when the electricity is needed.
A giant AI data center can suddenly add hundreds of megawatts of demand to a local grid. If the grid doesn't have enough generation or transmission capacity, somebody has to pay for:
new power plants
transmission lines
substations
distribution infrastructure
grid upgrades
additional reserve capacity
This is already happening. The EIA says data centers are driving much of the current growth in U.S. electricity demand, with particularly rapid growth in Texas and the PJM region. Its modeling found that unusually rapid demand growth could push Texas's 2027 wholesale electricity price 79% above its baseline forecast.
PJM, which serves much of the Mid-Atlantic and Midwest, is also facing rapidly increasing demand and has projected roughly 70 GW of additional demand by 2038, much of it from large new energy users such as data centers.
So when residents say:
"Why should my electric bill go up because Amazon/Google/Microsoft wants to build an AI factory?"
That is a legitimate policy question.
And there is an important distinction here: a data center doesn't inherently have to raise residential rates. It raises rates when the utility's rate structure or regulatory decisions allow some of the infrastructure costs to be shifted onto ordinary customers.
2. Water: also real, but much more geographically dependent
The water issue is somewhat different.
Data centers use water primarily because computers generate enormous amounts of heat. Traditional evaporative cooling can consume substantial quantities of water.
A recent USGS synthesis notes that a 100-MW data center can consume around 2 million gallons of water per day, although actual consumption varies enormously with the cooling system and climate.
And some of that water is genuinely consumed, rather than merely withdrawn and returned. Water evaporated during cooling doesn't go back into the local water system.
The problem becomes particularly serious when a data center is put somewhere like:
Arizona
Nevada
parts of Texas
Southern California
other drought-prone regions
rather than somewhere with abundant water.
A 2026 study found that data-center cooling could require an additional 697 million to 1.45 billion gallons of peak water capacity per day nationwide by 2030 under some scenarios. Importantly, the researchers emphasize that the burden would be highly concentrated in the communities hosting the facilities.
That's a crucial point.
The national average can be misleading.
Suppose data centers consume a relatively modest percentage of America's total water.
That doesn't mean much to a town whose water system is already operating near capacity and suddenly has a customer wanting hundreds of thousands or millions of gallons a day.
It's analogous to electricity:
Nationally: perhaps manageable.
Locally: potentially enormous.
3. But there is an important misconception about "ruining the water"
I'm more cautious about the claim that data centers generally make water "bad for humans."
That's not normally what happens.
The bigger concern is depletion and competition for water, rather than a data center routinely contaminating a community's drinking-water supply.
Cooling systems can produce wastewater containing concentrated minerals and treatment chemicals, and poorly managed discharge can create environmental problems.
But the much more common concern is:
"Is this facility consuming water that our community cannot afford to lose?"
That's a very different problem from:
"Is this facility poisoning our drinking water?"
The first is a significant and growing issue in some locations. The second should not be generalized to data centers as a whole.
4. The really interesting thing is that many of these problems are solvable
I don't think the appropriate response is:
"Stop building data centers."
Nor do I think the appropriate response is:
"Build as many as the technology companies want and let taxpayers deal with the consequences."
There is a much better middle ground.
I would establish a basic principle:
If a data center creates a substantial new infrastructure requirement, the data center—not existing residential customers—should pay the incremental cost.
That single principle could dramatically change the economics.
For example, suppose:
Company wants a 1-GW data center.
The utility determines that it requires:
$600 million in transmission upgrades
$300 million in substations
$400 million in additional generation
$100 million in other infrastructure
The company shouldn't simply be allowed to connect to the existing system and have those costs spread among millions of households.
The new load should pay its proportional share.
Pennsylvania's new rules are moving in this direction: the state is requiring data centers to address their power costs and environmental impacts more directly.
5. I would also require "full-cost" electricity contracts
This is perhaps the most important reform.
A data center should have a special electricity rate reflecting:
Generation + transmission + distribution + capacity + grid upgrades + reliability costs.
And there should be a requirement that the contract is long enough to protect the utility and its other customers if the data center shuts down.
Otherwise you can end up with a situation where:
Data center arrives → utility builds enormous infrastructure → data center later leaves → ordinary ratepayers are left paying for it.
That's backwards.
The risk should primarily belong to the company creating the extraordinary demand.
6. Require data centers to bring new power—not just consume existing power
This is another potentially powerful solution.
For very large facilities, I would require the developer to demonstrate where its electricity is coming from.
For example:
$600 million in transmission upgrades
$300 million in substations
$400 million in additional generation
$100 million in other infrastructure
The company shouldn't simply be allowed to connect to the existing system and have those costs spread among millions of households.
The new load should pay its proportional share.
Pennsylvania's new rules are moving in this direction: the state is requiring data centers to address their power costs and environmental impacts more directly.
5. I would also require "full-cost" electricity contracts
This is perhaps the most important reform.
A data center should have a special electricity rate reflecting:
Generation + transmission + distribution + capacity + grid upgrades + reliability costs.
And there should be a requirement that the contract is long enough to protect the utility and its other customers if the data center shuts down.
Otherwise you can end up with a situation where:
Data center arrives → utility builds enormous infrastructure → data center later leaves → ordinary ratepayers are left paying for it.
That's backwards.
The risk should primarily belong to the company creating the extraordinary demand.
6. Require data centers to bring new power—not just consume existing power
This is another potentially powerful solution.
For very large facilities, I would require the developer to demonstrate where its electricity is coming from.
For example:
Option A
Build or finance new solar/wind generation plus storage.
Build or finance new solar/wind generation plus storage.
Option B
Finance new nuclear generation.
Option C
Enter into long-term contracts for new generation and associated transmission.
Option D
Provide its own generation and storage sufficient to reduce its demands on the grid during emergencies.
The key is that:
Buying a certificate saying "we purchased renewable energy somewhere" shouldn't automatically qualify.
If a data center consumes enormous quantities of electricity from a constrained grid in Virginia while buying renewable-energy credits from a wind farm hundreds of miles away, that doesn't necessarily solve Virginia's capacity problem.
7. Make data centers interruptible during grid emergencies
This is an idea I particularly like.
A household can't easily be told:
"Turn off your refrigerator for three hours."
A hospital can't be told:
"Please shut down."
But an AI data center potentially has much more flexibility.
Some computing workloads can be shifted to another facility or scheduled at another time.
Therefore, large data centers could receive lower electricity rates in exchange for agreeing to reduce their load during grid emergencies.
PJM is already considering mechanisms involving large loads and emergency curtailment.
That could actually make data centers useful to the grid rather than merely a burden on it.
8. Water needs an equivalent system
I'd apply essentially the same principle to water.
Before approving a large data center:
The developer should have to disclose:
maximum gallons/day
average gallons/day
seasonal peak consumption
source of the water
percentage that is potable
percentage that is recycled
wastewater production
expected water consumption during drought
impact on the community's water capacity
And importantly:
Peak demand should matter—not merely annual consumption.
A water system can survive a data center consuming 500,000 gallons/day fairly comfortably if that demand is predictable.
But if it suddenly needs several million gallons on the hottest days of summer, that can require a completely different water infrastructure investment.
9. Don't use drinking water if you don't need drinking water
This is one of the easiest solutions.
Data centers can potentially use:
reclaimed wastewater
treated sewage effluent
industrial water
captured rainwater
closed-loop cooling
air cooling
hybrid cooling systems
rather than municipal drinking water.
Modern cooling technologies can dramatically reduce water consumption, and some new facilities can operate with little or no water for cooling.
The technology isn't the main obstacle.
Economics and regulation are.
If pristine municipal water is cheap, while installing a sophisticated recycling system is expensive, developers have an incentive to use the cheap water.
So government should change the incentives.
10. Here's the policy I'd favor
If I were designing a national framework, I'd establish something like a Data Center Community Impact Standard.
Any data center above a certain size—say 50 or 100 MW—would have to demonstrate five things before receiving approval:
Provide its own generation and storage sufficient to reduce its demands on the grid during emergencies.
The key is that:
Buying a certificate saying "we purchased renewable energy somewhere" shouldn't automatically qualify.
If a data center consumes enormous quantities of electricity from a constrained grid in Virginia while buying renewable-energy credits from a wind farm hundreds of miles away, that doesn't necessarily solve Virginia's capacity problem.
7. Make data centers interruptible during grid emergencies
This is an idea I particularly like.
A household can't easily be told:
"Turn off your refrigerator for three hours."
A hospital can't be told:
"Please shut down."
But an AI data center potentially has much more flexibility.
Some computing workloads can be shifted to another facility or scheduled at another time.
Therefore, large data centers could receive lower electricity rates in exchange for agreeing to reduce their load during grid emergencies.
PJM is already considering mechanisms involving large loads and emergency curtailment.
That could actually make data centers useful to the grid rather than merely a burden on it.
8. Water needs an equivalent system
I'd apply essentially the same principle to water.
Before approving a large data center:
The developer should have to disclose:
maximum gallons/day
average gallons/day
seasonal peak consumption
source of the water
percentage that is potable
percentage that is recycled
wastewater production
expected water consumption during drought
impact on the community's water capacity
And importantly:
Peak demand should matter—not merely annual consumption.
A water system can survive a data center consuming 500,000 gallons/day fairly comfortably if that demand is predictable.
But if it suddenly needs several million gallons on the hottest days of summer, that can require a completely different water infrastructure investment.
9. Don't use drinking water if you don't need drinking water
This is one of the easiest solutions.
Data centers can potentially use:
reclaimed wastewater
treated sewage effluent
industrial water
captured rainwater
closed-loop cooling
air cooling
hybrid cooling systems
rather than municipal drinking water.
Modern cooling technologies can dramatically reduce water consumption, and some new facilities can operate with little or no water for cooling.
The technology isn't the main obstacle.
Economics and regulation are.
If pristine municipal water is cheap, while installing a sophisticated recycling system is expensive, developers have an incentive to use the cheap water.
So government should change the incentives.
10. Here's the policy I'd favor
If I were designing a national framework, I'd establish something like a Data Center Community Impact Standard.
Any data center above a certain size—say 50 or 100 MW—would have to demonstrate five things before receiving approval:
1. Power neutrality
It pays the full incremental cost of generation, transmission, and grid upgrades attributable to it.
2. Water neutrality
It cannot consume more local water capacity than the community can sustainably provide without reducing existing users' reliability.
It cannot consume more local water capacity than the community can sustainably provide without reducing existing users' reliability.
3. Infrastructure neutrality
It pays for the infrastructure expansion that it causes.
4. Emergency flexibility
It must be capable of reducing electricity consumption during grid emergencies, unless it supplies sufficient independent generation.
5. Transparency
Electricity and water consumption must be publicly reported.
That last one is surprisingly important.
Research published in 2026 points out that data-center water-use reporting remains inadequate, making it difficult for communities to determine the actual consequences of proposed projects.
11. And I'd change the tax-break system
This may be the biggest political issue.
States have been offering enormous tax incentives to attract data centers because they bring investment and property-tax revenue.
But if a community gives a company hundreds of millions of dollars in tax benefits while simultaneously spending hundreds of millions upgrading the electric grid and water system for that company, the economic benefit can be much less impressive than it initially appears.
The better approach is:
No major tax subsidy unless the community can demonstrate a net economic benefit after accounting for infrastructure, electricity, water, environmental and public-service costs.
And the calculation should be public.
No secret agreements.
No nondisclosure agreements hiding the deal from residents.
No assumption that "jobs" automatically make everything worthwhile.
12. There is actually a fascinating opportunity here
The enormous electricity demand of AI could also accelerate America's energy infrastructure, if handled correctly.
The United States needs more:
transmission
nuclear power
solar
wind
geothermal
storage
efficient natural-gas generation
advanced grid management
anyway.
If data-center companies pay for a substantial portion of that new infrastructure, they could effectively become anchor customers financing America's next generation of electricity infrastructure.
That's potentially beneficial.
The danger is allowing the opposite arrangement:
Private company gets the economic benefits while the public absorbs the infrastructure costs.
That's the fundamental issue, in my view.
My bottom line
I'd rate the concerns this way:
Concern
How real?
Biggest problem
Huge electricity consumption
Very real
Rapid local demand growth
Higher electricity prices
Real in some regions
Infrastructure/capacity costs shifted to ratepayers
Grid reliability
Very real
Insufficient generation/transmission
Huge water consumption
Very real in some locations
Local water scarcity
Data centers "using up America's water"
Overstated nationally
Water problems are highly regional
Data centers making drinking water unsafe
Possible locally, but overstated as a general claim
Wastewater/discharge and depletion issues
Data centers inevitably harming communities
No
Depends enormously on location and regulation
Problem can be solved
Yes
Mostly through siting, pricing, technology and regulation
And there's one particularly important conclusion from the recent research:
We shouldn't treat "data centers" as one homogeneous thing.
A 20-MW facility in a water-rich area with closed-loop cooling and abundant generation is a completely different proposition from a 1-GW AI campus in a drought-stricken community whose electric grid is already constrained.
The location and infrastructure arrangement may matter more than the mere existence of the data center.
That's why I think the best policy isn't an outright national ban or moratorium. It is essentially "you can build it, but you must pay your own way and you cannot make the community less able to provide affordable electricity and water to its existing residents."
That principle is increasingly showing up in state policy because the backlash has become politically significant in 2026.
If you'd like, I can also dig into the other side of this question: exactly how much a typical household's electric bill could rise because of a large data center, who actually pays those costs, and which states are doing the best—and worst—job of protecting ordinary ratepayers.
