I'm glad you enjoyed it, thank you! I agree that desalination will likely become part of the solution, particularly for coastal regions. The challenge is that it's highly energy-intensive, expensive, and requires enormous infrastructure. More importantly, it does little to solve the problem for inland states like Arizona, where transporting desalinated water hundreds of miles becomes prohibitively costly. In many cases, the real constraint isn't the technology itself, but where water is available relative to where industry, agriculture, and population growth are concentrated.
i think this all happened because for the most part datacenter tech has been behind because it was "good enough" then ai demand just showed how unscalable it was all oversudden, but i think once the hype dies down the need for this much compute will die out and the problem will largely self regulate
Thank you for your comment! I get where you’re coming from, and I think that’s a reasonable way to look at it. My view is a bit different though. I don’t think this is just an AI hype cycle exposing outdated data center infrastructure. I think we’re entering a much more structural race for compute. I’m actually working on an article about exactly this, curious to hear what you think once it’s out!
The TSMC Phoenix frame is right - but the water constraint runs one layer upstream of the fab. The reagent chemistry that separates and refines the minerals going into that fab is hydrometallurgical: sulfuric acid leaching, solvent extraction cascades, precipitation and washing steps. All of it is water-intensive at industrial scale. Arizona produces 70%+ of US copper. Utah, Colorado, Nevada account for most of the rest. The processing buildout the US needs to reduce semiconductor dependence on Chinese inputs would have to be built in exactly the states facing the most severe structural water deficits. The Colorado River Compact update deadline is end of 2026 - the same window as DFARS and the critical minerals supply chain buildout. Water is the growth constraint the supply chain independence argument doesn't price in. The resource and the reagent chain are the same scarcity problem, two layers apart.
Although I am not knowledgeable enough to verify the details, I still find this comment great as it pinpoints to water dependencies that are not apparent in first check. Thanks
More a complex problem than a complicated one. Sure data centers provide for more jobs and tax income but those jobs involve people, and people require housing, energy, food, and water. "Qu'ils mangent de la brioche" isn't possible absent agriculture...
You've named the layer beneath the AI-as-geology story: the megawatt needs the liter. Arizona freezing 500,000 homes while approving the fab isn't hypocrisy — it's the state publishing its exchange rate, dollars per liter over people per liter. And once CME lists the scarcity, the doctrine of 1848 becomes collateral: Owens Valley again, with term sheets instead of dynamite. Superb essay.
The water mafia is for real, and they’ve been inconspicuously buying up rights to land with any water. Someday, not too far off into the future, the water business will rival the oil business.
Thank you for sharing this interesting article.
Surely, in wealthy countries at least, nuclear and solar powered desalination plants can fix this issue?
LF
I'm glad you enjoyed it, thank you! I agree that desalination will likely become part of the solution, particularly for coastal regions. The challenge is that it's highly energy-intensive, expensive, and requires enormous infrastructure. More importantly, it does little to solve the problem for inland states like Arizona, where transporting desalinated water hundreds of miles becomes prohibitively costly. In many cases, the real constraint isn't the technology itself, but where water is available relative to where industry, agriculture, and population growth are concentrated.
Thanks - that makes sense!
i think this all happened because for the most part datacenter tech has been behind because it was "good enough" then ai demand just showed how unscalable it was all oversudden, but i think once the hype dies down the need for this much compute will die out and the problem will largely self regulate
Thank you for your comment! I get where you’re coming from, and I think that’s a reasonable way to look at it. My view is a bit different though. I don’t think this is just an AI hype cycle exposing outdated data center infrastructure. I think we’re entering a much more structural race for compute. I’m actually working on an article about exactly this, curious to hear what you think once it’s out!
Given we have the Great Lakes, and those cities need investment, it's mad we're building data centers in the desert. But financial logic doesn't care.
Smart!!
The TSMC Phoenix frame is right - but the water constraint runs one layer upstream of the fab. The reagent chemistry that separates and refines the minerals going into that fab is hydrometallurgical: sulfuric acid leaching, solvent extraction cascades, precipitation and washing steps. All of it is water-intensive at industrial scale. Arizona produces 70%+ of US copper. Utah, Colorado, Nevada account for most of the rest. The processing buildout the US needs to reduce semiconductor dependence on Chinese inputs would have to be built in exactly the states facing the most severe structural water deficits. The Colorado River Compact update deadline is end of 2026 - the same window as DFARS and the critical minerals supply chain buildout. Water is the growth constraint the supply chain independence argument doesn't price in. The resource and the reagent chain are the same scarcity problem, two layers apart.
https://williamdavid.substack.com/p/the-chokepoint-guide-to-critical-133
Although I am not knowledgeable enough to verify the details, I still find this comment great as it pinpoints to water dependencies that are not apparent in first check. Thanks
More a complex problem than a complicated one. Sure data centers provide for more jobs and tax income but those jobs involve people, and people require housing, energy, food, and water. "Qu'ils mangent de la brioche" isn't possible absent agriculture...
You can transport the food in.
You've named the layer beneath the AI-as-geology story: the megawatt needs the liter. Arizona freezing 500,000 homes while approving the fab isn't hypocrisy — it's the state publishing its exchange rate, dollars per liter over people per liter. And once CME lists the scarcity, the doctrine of 1848 becomes collateral: Owens Valley again, with term sheets instead of dynamite. Superb essay.
The water mafia is for real, and they’ve been inconspicuously buying up rights to land with any water. Someday, not too far off into the future, the water business will rival the oil business.
https://www.cfr.org/backgrounders/water-stress-global-problem-thats-getting-worse. It is a global problem that will create more refugees, which the more fortunate countries are beginning to obstruct (tragically).