They Said My Tesla Was Running on Coal

I bought a Tesla in 2013. Model S. This was before the current complications, when Elon Musk was still primarily famous for founding two companies that did extraordinary things almost everyone said couldn't be done.

The first thing people said when they found out:

"You know the electricity comes from coal, right? You're just moving the pollution."

I heard this dozens of times. It was delivered with a confidence that implied research. It hadn't involved any. The arithmetic was available to anyone with a high-school physics background and fifteen minutes, and it was decisive: even on a grid powered entirely by coal, an electric car was more efficient than a gasoline engine. A coal plant is inefficient at generating electricity — but an electric motor puts roughly 90% of that electricity into motion. A gasoline engine puts about 30% of its fuel into motion. The coal argument was technically true in one narrow, irrelevant way. It was wrong in every way that mattered.

It was also, notably, everywhere at once. Same talking point. Same framing. Same selective arithmetic. Same proposed conclusion: don't buy an electric car.

I am watching the same thing happen with AI and water.

The Structure of a Physics-Free Argument

Before getting to the numbers, it's worth naming the structure, because it recurs.

A coordinated narrative of this kind has a few tells. It leads with an absolute number that sounds alarming without context: AI data centers consumed 264 billion gallons of water in 2025 — equivalent to the annual water needs of 1.8 million Americans. It omits the comparison that would immediately defuse the alarm. It conflates two things that are not the same thing. It proposes a solution that maps cleanly onto stop using the technology in question. And it spreads with a velocity that suggests something other than organic curiosity.

I am not making a claim about who coordinates these things. I am making a claim about pattern recognition. When you've seen the structure deploy twice, you start seeing it faster.

What "Water Usage" Actually Means

In a cooling tower — which is how most data centers manage heat — water is used to carry thermal energy away from servers. Some of that water evaporates. This is the "consumption" that gets reported.

Here is what doesn't get reported: evaporated water returns to the water cycle. It leaves the cooling tower as vapor and eventually falls as precipitation somewhere. It has not been destroyed. It has not been poisoned. It has not been removed from the hydrological system. The distinction between water withdrawn and water consumed is well-established in water science; conflating them is not an honest error, because the distinction is not subtle.

The alarming numbers that circulate also tend to blend what researchers call Scope 1 (the water your data center uses directly), Scope 2 (the water used by the power plants generating your electricity), and occasionally Scope 3 (supply chain). These categories are legitimate accounting tools when labeled. When they're added together and reported as a single figure with no breakdown, they're doing different work. The power plants generating grid electricity would be consuming cooling water regardless of what the electricity is used for — attributing all of that water to AI specifically requires a level of deliberate accounting creativity that doesn't happen by accident.

The Hamburger Problem

The scale comparison is where the argument should collapse under its own weight, but only if you do it.

A single hamburger requires approximately 5,000 liters of water to produce — mostly in the agriculture required to raise cattle. A cotton t-shirt requires around 2,700 liters. A single kilogram of steel requires in the vicinity of 235,000 liters.

A ChatGPT query, on-site, consumes roughly 2 to 25 milliliters of water, depending on the accounting methodology. The 25-milliliter figure — about a capful — uses the most expansive scope.

If you're genuinely concerned about freshwater availability, you are looking at a landscape where agriculture accounts for approximately 70% of global freshwater withdrawals, industry accounts for roughly 19%, and AI data centers represent a fraction of a fraction of that 19%. You are also looking at a landscape where billions of hamburgers are consumed annually with minimal public concern about hydrological crisis. The people promoting AI water anxiety are not, by and large, also promoting hamburger water anxiety, which tells you something about whether this is fundamentally about water.

The One Part That's Real

There is a legitimate concern buried in this narrative, and it deserves to be separated from the rest.

Data centers built since 2022 have been disproportionately sited in water-stressed regions — parts of Texas, Arizona, the Colorado River Basin. Evaporative cooling is most efficient in dry climates, so the geography makes engineering sense. But placing large industrial water consumers in areas experiencing long-term drought stress creates real local strain. Google's data center in The Dalles, Oregon, reportedly accounts for 29% of the city's total water consumption.

This is a siting and permitting problem. It is the kind of problem that gets addressed through disclosure requirements and local regulatory decisions — and some jurisdictions are already doing this. It is not evidence that AI is drinking the world's water. Conflating the two is how a legitimate policy concern gets weaponized into a physics-free global narrative.

The Pattern

In 2013, the coal argument was trying to get you to not buy an electric car. The electric car, on the actual arithmetic, was better for the environment than the gasoline alternative, and was going to get better as grids decarbonized — which they have.

Here is the part that I find most instructive: I am still hearing the coal argument. Right now. In 2026, with gasoline at four to five dollars a gallon and a market full of electric vehicles that didn't exist when I bought mine — the argument is still running. Unchanged. "You know the electricity comes from coal, right?"

This is not how organic misconceptions behave. A genuine misunderstanding, confronted with enough contrary evidence and economic reality, fades. People update. The coal argument has been comprehensively refuted by physics, by grid data, by the lived experience of millions of EV owners, and by a fuel price that makes the comparison academic — and it is still being repeated, with the same confidence, in the same words. That is not a misconception. That is infrastructure.

The water argument is trying to get you to something similar. I'll let you work out what.

The tell, both times, is the missing comparison. An argument that gives you an absolute number without telling you what to compare it to is not informing you. It is managing you. High-school physics gives you the tools to notice. Whether enough people were taught to use them is a separate, considerably darker question.