Voters approve 40 of 41 ballot measures to restrict data centers, CNBC says

Voters approved 40 of 41 proposed ballot measures restricting some planned data centers this week, CNBC reported November 5. Water and electricity topped residents’ concerns. Yet the local demand behind those concerns remains hard to measure: operators typically publish companywide usage figures, and cooling systems can trade lower water consumption for higher power demand.

The measures would place restrictions on some of the more than 2,000 new facilities planned across the country, according to CNBC. That is a vote for limits on development, not a finding that every proposed facility has the same resource footprint. Hardware, location and the cooling method all affect how much water a site needs.

Recirculating water does not remove the heat

The first distinction is between cooling the computers and cooling the facility. Older data centers often use fans to move heat away from lower-power servers. Modern AI facilities pack in high-power graphics processors, or GPUs, that produce much more heat. Liquid cooling carries that heat away from the equipment.

Most large AI data centers use a closed-loop liquid system, circulating the same water with little daily consumption. But that loop still has to shed the heat it collects. A separate cooling process transfers it to the outdoor environment—and that is where substantial water consumption can occur.

Dry cooling blows air across heat exchangers. When outside air is hot, refrigeration may be needed, increasing electricity demand. Evaporative cooling instead lets some water evaporate and carry heat away. It uses relatively little power, but can lose large quantities of water on the hottest days.

The local footprint gets lost in company totals

Operators also combine the two approaches. Equinix’s San Jose facility uses dry cooling in colder months and evaporative cooling when temperatures rise. CNBC visited its closed-loop system serving eight Nvidia Grace Blackwell racks owned by AI marketing startup Alembic. Equinix provides cooling, power and connections while customers rent space for their equipment.

Equinix does not disclose individual sites’ water consumption. Across its operations, it consumed 1.2 billion gallons in 2025, up 10% from the previous year. That companywide total does not tell a San Jose resident how much the nearby facility consumes—a gap shared by operators that report aggregate figures rather than site-level use.

A peak-day number is not an everyday forecast

Scale estimates need similar care. Shaolei Ren, a data-center researcher at the University of California, Riverside, told CNBC that a one-gigawatt AI facility can consume up to 10 million gallons daily. That is an upper-end estimate, not a standard daily requirement: the equipment, climate and cooling design change the result.

Meta’s planned one-gigawatt facility in Lebanon, Indiana, illustrates the difference. The company says its closed-loop system will use no water for most of the year. Yet the site’s water allocation could reach 8 million gallons a day by 2031, according to its law firm. Meta told CNBC it expects to reach that maximum only a few days annually.

On hotter days, Meta plans to spray mist to cool the air, reducing reliance on refrigeration for its dry-cooling system. The design shows why neither a closed-loop label nor a maximum water allocation alone describes a facility’s everyday demands. Weather determines when the external system needs extra help—and which resource supplies it.