Tech giants pouring an estimated $1 trillion into AI infrastructure are consuming far more water than their sustainability reports acknowledge. Microsoft, Google, and Amazon disclose only on-site data-center usage, ignoring the massive volumes evaporated at power plants generating their electricity. Meta stands alone among peers by tallying both direct and indirect consumption, a gap no U.S. law currently requires companies to bridge.

The disparity is staggering. Lawrence Berkeley National Laboratory found that indirect water consumption for U.S. data centers historically runs 12 times higher than direct use. Google's latest report illustrates the blind spot: the company logged 10.9 billion gallons in 2024, a 34% jump, nearly all for cooling. But that figure excludes water burned off at coal, nuclear, or gas plants feeding those facilities.

Power source dictates the hidden toll. Coal and nuclear plants demand vast cooling water; natural gas requires less; solar and wind need virtually none. As AI drives the largest U.S. infrastructure buildout in decades, experts warn that unreported demand could spark regional battles over increasingly scarce supplies. The accounting gap masks the true environmental footprint of the compute boom.

Investors and regulators now face a clear choice: accept voluntary disclosures capturing a fraction of reality, or mandate full-scope reporting before water conflicts escalate. The $1 trillion buildout proceeds regardless of whether the accounting catches up, and the regions hosting these facilities will bear the consequences.