A look at Google's US data center water usage, which it tries to keep secret: an estimated 2.3B gallons in three states in 2019, some with water shortages
In August 2019, the Arizona Municipal Water Users Association built a 16-foot pyramid of jugs in its main entrance in Phoenix.
Context & Ripple Effects
In 2020, Bloomberg had to estimate what Google would not say: roughly 2.3 billion gallons of water consumed by its US data centers in just three states, some of them short on supply — a figure the company declined to confirm even as the Arizona Municipal Water Users Association staged a 16-foot pyramid of jugs in Phoenix to dramatize local scarcity. Secrecy was the default posture for hyperscale water use at the time.
The arc since then runs toward forced disclosure: drought-stricken towns began pushing back, FOI requests pried open Virginia's consumption figures, and by mid-2026 both Google and Amazon were publishing their own numbers — Google pledging to be water positive by 2030, Amazon reporting ~2.5 billion gallons used in 2025 with a per-kWh intensity metric. This story is the baseline against which that reversal reads.
First-order effects
- Communities hosting Google's facilities in water-stressed states are making siting and expansion decisions without knowing the operator's actual draw, while local water authorities like the Arizona association absorb the political heat for allocations they cannot verify.
- Google's refusal to disclose leaves it exposed: every later estimate by journalists or FOI filings sets the public number instead of the company.
Second-order effects
- Rivals are pushed into disclosure arms races — Amazon now publishes annual gallon totals and per-kWh intensity, converting what was once a liability into a competitive talking point.
- FOI requests become the standard procurement tool for host municipalities, as Virginia's experience showed when filed documents revealed consumption there had grown well past its 2019 level.
Third-order effects
- Water is hardening into a utility-grade metric for AI infrastructure: operators report it like electricity, and 'water positive' pledges become the reputational floor rather than a differentiator.
- Siting economics shift structurally — long-term water availability joins power and fiber as a gating factor for where hyperscale campuses get approved, giving arid-region municipalities real veto leverage.
The trend: Data center water consumption is moving from guarded trade secret to standardized public disclosure, with AI buildout pressure turning local water politics into a constraint on hyperscale expansion.