Love how these types of posts conveniently exclude any of the additional context from the literal article they cite. Chasing engagement. Sounds like a big, scary number.
Context:
-Amazon's data centers use 0.12 L/kWh of water. Industry average: 0.84 L/kWh. 7x more efficient.
-vs. other hyperscalers: Microsoft FY25 WUE: 0.27 L/kWh. Meta 2024: 0.19 L/kWh. Amazon 2025: 0.12 L/kWh. Google's LLM-supporting data centers hit 1.15 L/kWh (though that is a different reporting boundary). Amazon's fleet-wide average is 0.12.
-Water intensity down 52% since 2021. Northern Virginia (largest IT load region) specifically: -42% YoY while demand grew.
-90% of the time, Amazon's data centers use zero water cooling = no water-based cooling. Water kicks in when temps exceed ~85°F.
-Amazon returned 3 gallons for every 4 used in 2025 (replenished / returned via projects). 50 replenishment projects coming online will return 5.8B gallons/year = >2x current withdrawals (obvious caveat is water impact is local vs. replenishment often compared to global)
For reference:
-Data centers as a share of global industrial water use: <0.5%. Amazon is a fraction of that fraction.
-U.S. lawn / landscape irrigation withdrawals are ~8B gallons/day.
-U.S. public water supply is 39B gallons/day.
-U.S. thermoelectric power withdrawals are 133B gallons/day (much of which is returned, btw; Withdrawals ≠ consumption)
So yes, water matters. Especially locally. But the honest critique is local watershed stress and disclosure boundaries, not pretending Amazon’s global data-center water footprint is remotely comparable to agriculture, lawns, public supply, or power-sector withdrawals.
Amazon said its data centers used 2.5 billion gallons of water worldwide last year, or about 5% of the amount metro Seattle consumes annually