By George Pullen, Chief Economist | Ai-CaRM, MilkyWayEconomy
The Illusion of Dry Intelligence
There is a quiet lie embedded in every AI invoice. It tells you that intelligence is cheap…that for a few cents per thousand tokens, you can rent cognition, scale it, industrialize it. That thinking, once scarce and expensive, has finally been democratized into a utility. But the invoice is incomplete. Because it does not list the most important line item: water.
We speak about AI as though it lives in the cloud…weightless, abstract, detached from the physical world. It does not.AI lives in data centers. Data centers live on land. Land exists within ecosystems. And ecosystems, whether we respect it or not, run on water. Every token you generate is heat. Every unit of heat must be removed. And in many of the world’s most advanced data centers, that heat is removed with water…evaporated, circulated, consumed. You are not buying intelligence. You are participating in a thermodynamic transaction.
Tokens Are a Resource Contract
The industry has trained us to think of tokens like software pricing…clean, scalable, predictable. But tokens behave less like software and more like electricity futures, water rights, and industrial inputs. Because that is what they are built on. When you scale AI usage, you are not just increasing compute. You are increasing energy demand, cooling demand, and water consumption. And unlike code, those inputs do not scale infinitely. They compete.
The Coming Collision
The United States is already experiencing water stress across multiple regions. The same regions that have become attractive for data center expansion…cheap land (Texas), favorable tax structures (Texas), proximity to infrastructure (Texas)…are often the same regions where water is becoming constrained. This is not theoretical. It is structural. You now have three systems competing for the same resource: cities that need drinking water, agriculture that feeds populations, and data centers that enable machine cognition. Only one of these is new. And only one is currently priced as though its inputs are infinite.
The Externality That Will Not Stay External
Today, water costs are buried. They are abstracted into total infrastructure cost…smoothed across pricing models…hidden behind the elegance of per-token billing. But externalities have a way of becoming explicit. When they do, they do not arrive gradually. They arrive as shocks…usage restrictions, tiered pricing spikes, regulatory intervention, geographic limitations on expansion. And when that happens, the illusion collapses. Tokens will no longer feel cheap. They will feel contested.
The Geography of Thought
We are approaching a world where not all thinking costs the same. Where a token generated in a water-rich region is cheaper, and a token generated in a drought-stricken region carries a premium. This introduces a new form of inequality: cognitive geography. Where your infrastructure lives will determine how affordably you can think with machines. And organizations will be forced to make decisions not just about cloud providers…But about watersheds.
The Inversion No One Modeled
For years, the assumption has been simple: machines are cheaper than humans. That assumption holds… until it doesn’t. Because when the cost of supporting machine cognition rises—through energy constraints, water scarcity, or regulatory friction—the equation shifts. Not because humans became more efficient. But because the system required to sustain artificial cognition became more expensive than expected.There will be environments… industries…moments…Where it is simply cheaper to use a human. Not because they are better. But because they are less resource-intensive.
The Strategic Blind Spot
Most organizations are not modeling this. They are tracking token spend the way they tracked cloud spend a decade ago… as a cost optimization problem. It is not. It is a resource dependency problem. A continuity problem. A geopolitical problem. Because when your ability to think is tied to constrained physical resources, you have not eliminated risk. You have relocated it.
The Question That Matters
The industry is still asking:
“How much can AI replace?”
It should be asking:
“What does it cost the world to support this level of machine thinking?”
Because that cost is not just financial. It is environmental. It is infrastructural. It is shared.
The Final Line
There is a sentence that will sound strange today, and obvious in hindsight: We did not run out of intelligence. We ran into the limits of what it takes to produce it. And when that realization sets in… The price of a token will no longer be measured in cents. It will be measured in resources you can no longer ignore.
George