The semiconductor water crisis nobody wants to

I’ve been tracing the water supply chains for five major semiconductor fabs. The picture is precarious.

TSMC in Hsinchu, Taiwan: 197,000 tons of water per day. Taiwan had a severe drought in 2021 that forced water rationing across the island. TSMC got priority, because chips are Taiwan’s most important export. Farmers got their water cut. That’s the trade-off nobody advertises.

TSMC in Arizona: roughly 4.5 million gallons per day. In a desert. Arizona is in a 23-year drought. The Colorado River, which supplies much of the state’s water, is at historically low levels. TSMC is recycling heavily and has invested in water treatment, but the fundamental equation is: water-intensive manufacturing in a water-scarce place.

Intel in Ohio: projected 5 million gallons per day when fully operational. Ohio has more water than Arizona, but the local water infrastructure wasn’t designed for this kind of industrial demand. Upgrades are underway. The costs are measured in hundreds of millions.

Samsung in Pyeongtaek, South Korea: exact figures aren’t public, but industry estimates put it at comparable levels to TSMC’s main fabs. Korea’s been investing in water recycling technology aggressively.

ASML in Veldhoven, Netherlands: less water-intensive (they build the machines, not the chips), but their testing and calibration still requires ultra-pure water in quantities that surprised me when I looked into it.

Why chips need so much water

Semiconductor manufacturing requires ultra-pure water. Not just clean. Pure. Fewer than 5 parts per billion of contaminants. Regular tap water has roughly 200-500 parts per million. The purification process itself uses water. And the chips need to be rinsed at multiple stages of the lithography process.

A single 300mm silicon wafer goes through about 2,000 gallons of ultra-pure water during processing. TSMC processes hundreds of thousands of wafers per month.

The math is enormous. And every new node (5nm to 3nm to 2nm) requires more processing steps, which means more water per wafer.

The climate collision

Climate change is making water scarcer in many of the places where fabs exist or are being built. Taiwan’s drought risk is increasing. Arizona is structurally water-stressed. Even water-rich regions are facing allocation conflicts as population growth competes with industrial demand.

The AI boom is making this worse. Every new data center needs chips. Every new chip needs water. The demand for AI computation is growing exponentially. The water supply isn’t.

I pulled data from the World Resources Institute and the US Geological Survey. The overlap between “regions with water stress” and “regions with fabs or planned fabs” is uncomfortably large.

What I’m not seeing

I’m not seeing this conversation in the mainstream tech discourse. GPU counts, yes. Training costs, yes. Energy consumption, starting to emerge. Water? Almost never.

It’s like talking about a car’s engine without mentioning fuel. The future of computing depends on silicon, electricity, and water. Two of those get discussed. The third, the one that’s actually scarce, gets ignored.

I don’t have a solution. I don’t think there’s a simple one. Water recycling helps but doesn’t eliminate consumption. Desalination is energy-intensive. Moving fabs to water-rich regions introduces other constraints.

But ignoring the problem doesn’t make it smaller. And right now, the semiconductor industry’s water dependency is the biggest underreported risk in technology.

Related thinking: