AI Is Not the Water Villain

What we miss when we blame the new tool and ignore the old waste
By Brandon Askew - Founder, Application Developer & AI Builder

The pattern I keep seeing

I have been building software long enough to see this movie a few times.

Every time a new technology shows up and starts to move society, we panic. Not always for bad reasons. Sometimes the concerns are real. But the pattern is real too: we take the new thing, isolate its footprint, and make it carry the full moral weight of systems that were already wasteful before the new tool arrived.

I saw it with the early consumer internet. People said it would break human connection. I saw it with cloud computing. People looked at server farms and said the grid could not handle it. Then ChatGPT showed up, and for me it was obvious right away that something had changed.

This was not just another app. This was a new layer of work.

And right on schedule, the same cycle showed up again.

This time the headline is water. AI is drinking the watersheds dry. Data centers are evaporating the future. Every prompt has a hidden environmental bill attached to it.

Here is the thing: I am not saying the footprint is fake. Data centers are real buildings. They use power. They use water. They require chips, cooling, land, permits, and infrastructure. We should talk about all of that honestly.

But my problem is with the way the conversation gets framed.

We count the cost of the new tool under a microscope, while the old systems it could replace are treated like they have no cost at all.

The blind spot: we treat the old system like it is free

There is a name for this: status quo bias.

In normal language, it means we get used to the mess we already live with. If a system has been wasting resources for decades, it starts to feel normal. It becomes background noise. We do not feel the cost anymore because the cost has been spread across habits, buildings, roads, fuel, packaging, supply chains, paperwork, and workflows we stopped questioning a long time ago.

Then a new technology shows up with measurable numbers, dashboards, cooling towers, GPU specs, and clean accounting. Suddenly every gallon is visible. Every watt is counted. Every prompt becomes something people can blame.

That is not an honest comparison.

It is not enough to ask, “What does AI use?”

We also have to ask, “What system is AI replacing, shrinking, or optimizing?”

If we do not ask both questions, we are not doing environmental analysis. We are just defending whatever already exists.

The numbers change when you compare real systems

The public conversation usually focuses on the fully loaded footprint of AI: mining the materials, making the chips, building the data center, training the model, cooling the servers, and running the query.

Fine. Count it all.

I actually like that approach because it forces honesty.

But then we need to count legacy systems the same way. Not just the gallon of gas in the tank. Count the car manufacturing. Count the fuel refining. Count the supply chain. Count the physical movement of stuff. Count the building, shipping, storage, waste, and disposal.

When you do that, the AI numbers do not look like the villain. They look like a visible new cost sitting next to a much larger old cost that most people stopped seeing.

These numbers are not meant to be perfect moral scorecards. They are scale checks. They depend on assumptions, and AI water estimates vary by model, location, cooling system, energy source, and whether you count only runtime or a broader lifecycle footprint.

But the direction of the comparison matters.

The exact number will move depending on the source and the model. That is why the better question is not, “Can I find a scary number?”

The better question is, “Compared to what?”

Amortized infrastructure is different from repeated physical waste

This is the part that gets missed over and over again.

A frontier AI model may take a large upfront resource investment to train. That matters. Model development, data center construction, hardware manufacturing, energy use, water use, and end-of-life disposal should all be part of the conversation.

But once that model exists, it becomes a digital asset that can be used by millions or billions of people. The upfront cost gets spread across a massive number of users and tasks.

That is not how most physical systems work.

A hamburger does not get shared across a billion people. A gallon of gasoline does not get reused after it is burned. A plastic syringe does not become more efficient because more people used the hospital that day. Physical waste has to be manufactured, shipped, used, cleaned up, or thrown away again and again.

So when someone says, “AI training used millions of gallons of water,” the honest follow-up is:

Over how many users?
Over how many tasks?
Replacing what older workflow?
Over what time period?
In what location?
Using what cooling method?
Powered by what energy source?

Without that context, the number may be emotionally powerful, but it is not very useful.

Shifting code versus moving mass

Think about the old physical media world.

To watch a movie, we used to manufacture plastic VHS tapes and DVDs. We printed cases. We boxed them. We shipped them on trucks. We stocked them in stores. Then people drove to rent or buy them. When the format became old, huge amounts of that plastic went to landfills.

At the time, that felt normal.

Nobody thought about the full physical lifecycle every time they watched a movie. It was just how the system worked.

Streaming did not become perfect. It still uses electricity, networks, servers, and devices. But it compressed a giant physical process into data moving through cables and networks.

That is the bigger pattern.

Over time, technology keeps trying to move us from heavy physical systems toward lighter digital systems. Not always cleanly. Not automatically. Not without tradeoffs. But the direction matters.

AI is part of that same shift.

It compresses certain kinds of work into math: research, drafting, coding, planning, diagnostics, logistics, paperwork, analysis, and coordination.

The question is not whether that math has a footprint. It does.

The question is whether that footprint is smaller than the physical and administrative waste it can remove.

The Star Trek lens: when the normal technology has a cost

There is a Star Trek: The Next Generation episode, “Force of Nature,” that keeps coming to mind here.

It is not about AI. It is about warp travel.

In the episode, the question is not whether warp drive is useful. It clearly is. Warp is the infrastructure of exploration, trade, diplomacy, rescue, and everyday movement across the Federation. The question is what happens when a technology that became normal starts showing environmental consequences that were easy to ignore while the system was working.

That is the part that connects to this piece.

Once the impact becomes visible, the question changes from, “Is this technology valuable?” to, “Can we keep using it the same way without accounting for the damage?”

That is where I think the AI water conversation should go.

Not denial. Not panic. Better accounting.

The hard part is not admitting that AI has a footprint. It does. The hard part is refusing to let familiar systems stay invisible just because we already built our lives around them.

Healthcare shows the double standard clearly

Healthcare is a hard example because the mission is important.

Medicine protects life. Hospitals matter. Doctors, nurses, and care teams do work that society depends on.

But that does not mean the system gets to be invisible.

The U.S. healthcare sector is responsible for roughly 8.5% of the country’s greenhouse gas emissions. Hospitals are also among the most water-intensive commercial buildings in the country. The exact water number varies by source and facility: EPA/ENERGY STAR Portfolio Manager data has reported a hospital median around 315 gallons per bed per day, while other healthcare facility discussions commonly cite figures around 570 gallons per staffed bed per day.

Either way, one hospital bed, for one day, can equal the water footprint of a very large amount of AI use under lower amortized query estimates. A one-week stay can approach roughly 2,200 to 4,000 gallons of direct water use.

Graphic reference: The Institutional Water Debt: U.S. Healthcare graphic should be treated as a scale-check visual, not a fixed universal audit. It is most useful because it breaks the bed-day footprint into familiar buckets - patient water, laundry, sterilization, cooling, kitchen use, sanitation, and supply chain. That is exactly the kind of cost we stop seeing once it becomes normal infrastructure.

Again, I am not saying hospitals are bad.

I am saying our mental accounting is broken.

We accept water-heavy, energy-heavy, supply-chain-heavy institutions because they are familiar and necessary. But when a data center is built to run medical analysis, automate paperwork, support diagnostics, improve logistics, or reduce administrative waste, the headline often becomes: AI is the environmental threat.

That is too simple.

It lets the old waste hide behind familiarity while the new tool takes the blame.

AI should not get a free pass either

None of this means AI companies should be allowed to build wherever they want, consume whatever they want, and call it innovation.

That is not the point.

If a data center is built in a dry region already struggling with water, that is a real problem. If local residents are competing with cooling infrastructure for water, that is a real problem. If companies are vague about energy, water, offsets, cooling methods, or long-term local impact, that deserves scrutiny.

But that is a site planning problem. A zoning problem. A utility problem. A cooling design problem. A corporate accountability problem.

It is not proof that asking an AI model a question is some new personal climate sin.

The better questions are:

Where is the data center?
How is it cooled?
What power source supports it?
What local resources are being stressed?
What old workflow is the compute replacing?
What physical waste is being reduced?
What paperwork, travel, shipping, duplication, or supply-chain drag is being removed?

That is the conversation we should be having.

The useful criticism

The easy headline is: AI uses water.

The honest headline is: everything uses resources, and we need to compare systems fairly.

If we only count the cost of the future and ignore the cost of the past, we are not protecting the planet. We are protecting the status quo.

That is what bothers me about so much of the AI environmental conversation. It is not that people care about water. We should care about water, especially in places where local resources are limited and communities are directly affected.

What bothers me is the selective outrage.

A prompt becomes a moral failure, while the old systems burning fuel, moving plastic, washing linens, manufacturing disposables, and pushing paperwork across huge supply chains get treated as normal life.

I do not want AI to be above criticism.

I want the criticism to be useful.

Useful criticism does not stop at panic. It asks better engineering questions. It looks at full lifecycle costs. It compares old and new systems honestly. It pushes companies to build smarter, in better places, with better cooling, better energy, and real accountability.

The lesson: AI does have an environmental footprint. But blaming the new tool while ignoring the old waste is not sustainability. It is status quo bias with better headlines.

Because shifting code is almost always going to be lighter than moving mass.

The work now is making sure we build that digital future responsibly instead of pretending the physical past was clean.

‹

›

FAQ

Does AI use water?

Yes. AI runs on data centers, and data centers use water directly for cooling and indirectly through the electricity systems that support them. The amount depends on the model, location, cooling method, energy source, and how the estimate is calculated.

Is the concern about AI water use real?

Yes, especially when data centers are built in water-stressed areas or when companies are not transparent about local impacts. The concern is real. The problem is when the conversation stops at panic instead of comparing AI to the older systems it may replace or reduce.

What is status quo bias?

Status quo bias is the habit of treating existing systems as normal, even when they are wasteful. In this case, it means AI gets blamed because its footprint is visible, while older systems like transportation, physical media, food production, healthcare, and paperwork are treated as background.

Why compare AI to cars, food, streaming, and hospitals?

Because environmental impact should be compared across real systems, not judged in isolation. The point is not that AI is free. The point is that every system uses resources, and we need to ask what a new technology is replacing, shrinking, or optimizing.

Should AI companies still be held accountable?

Absolutely. Data center location, energy source, cooling design, water stress, transparency, and community impact all matter. Fair comparison is not a free pass. It is a better starting point for useful criticism.