Data center water consumption has become the invisible crisis of AI’s boom. More than a dozen shareholder resolutions filed ahead of 2026 annual meetings are demanding that Amazon, Microsoft, and Alphabet’s Google disclose exactly how much water and power their U.S. data centers consume. The timing is not coincidental: North American data centers used nearly 1 trillion liters of water in 2025, a figure that rivals the annual water demand of New York City.
Key Takeaways
- Investors filed 12+ resolutions pressuring Amazon, Microsoft, and Google for U.S. data center water and energy transparency.
- North American data centers consumed nearly 1 trillion liters of water in 2025, equivalent to NYC’s annual demand.
- U.S. data centers used 17 billion gallons directly for cooling in 2023, with indirect electricity-related usage reaching 211 billion gallons.
- Data center water consumption is projected to double globally by 2030 as AI workloads surge.
- Meta’s water usage rose 51% from 3,726 megaliters in 2020 to 5,637 megaliters in 2024.
Why Big Tech is Hiding Data Center Water and Power Numbers
The three tech giants report sustainability metrics in wildly different ways, making true comparison impossible. Meta discloses water usage for owned sites only, excluding leased facilities and under-construction data centers. Google provides the most granular data, breaking down consumption by owned and leased sites, but excludes third-party operated facilities. Microsoft reports total operational water usage without any breakdown by data center or facility type. Amazon goes further than Microsoft but still avoids publishing total consumption figures, instead reporting water usage per unit of power generated.
This fragmentation is by design. When one Iowa data center consumed 1 billion gallons—enough to supply all of Iowa’s residential water needs for five days—in a single year, the lack of transparency meant local communities discovered the impact only after the fact. Investors argue this opacity masks the true environmental cost of the AI race. Giovanna Eichner, a shareholder advocate at Green Century Capital Management, framed the issue clearly: the industry must ensure that short-term AI gains do not come at the cost of long-term climate and financial risk.
The Scale of Data Center Water and Power Consumption in the U.S.
The numbers reveal why investors are alarmed. U.S. data centers drew 17 billion gallons of water directly for cooling in 2023, according to the Lawrence Berkeley National Lab. But that is only the direct consumption. Indirect water usage via electricity generation totaled 211 billion gallons that same year. Combined, that is 228 billion gallons annually, and the trajectory is steep. Electricity demand for U.S. data centers is projected to reach 1,050 TWh by 2030, up from 176 TWh in 2023—a 128% increase in less than a decade.
The largest facilities consume staggering amounts. A single Google data center in Storey County, Nevada used 1.5 million gallons in 2023. A more modest Google facility in Pflugerville, Texas—air-cooled and therefore more efficient—consumed 10,000 gallons, roughly equivalent to two months of water for a typical Texas home. Across the entire U.S., approximately 5,426 data centers consume an estimated 449 million gallons per day, or 163.7 billion gallons annually. The largest facilities rival the water consumption of towns with populations of 10,000 to 50,000 people.
AI Is Accelerating Data Center Water and Power Consumption Growth
The crisis is not static—it is accelerating. Global data center electricity consumption is projected to rise 128% by 2030, driven almost entirely by AI training and inference workloads. Water demand is expected to double by 2030 as well. Yet fewer than one-third of data center operators even tracked water consumption in 2016, meaning the industry has historically treated the resource as abundant and invisible.
The problem compounds when considering how data centers source their electricity. Fifty-six percent of U.S. data center power comes from fossil fuels, with coal being the most water-intensive option at 19,185 gallons per megawatt-hour. This means a data center powered primarily by coal consumes water twice over: once directly for cooling, and again indirectly through the water needed to generate its electricity.
Microsoft, Amazon, and Google have all committed to renewable energy targets, but the scale of growth outpaces the speed of transition. Amazon’s 2025 sustainability report frames the company as increasingly disclosing site-specific water consumption data and committed to being a good neighbor through efficiency investments and new energy sources. Yet without mandatory, standardized reporting, these pledges remain unverifiable and subject to corporate interpretation.
What Investors Are Demanding—and Why It Matters
The shareholder resolutions are straightforward: disclose total U.S. data center water and power consumption, broken down by facility and by year. This would allow investors to assess climate and financial risk, and would enable communities to advocate for their own water resources. The requests are not anti-technology; they are pro-transparency. Without it, the industry can optimize for profit while externalizing environmental costs onto drought-stricken regions in the U.S., Chile, Uruguay, and Sweden.
Green Century Capital Management is in discussions with Nvidia on an AI-climate resolution as well, signaling that the pressure will extend beyond hyperscalers to the chip makers that power them. If major investors force disclosure, they also force accountability. And accountability forces innovation toward water-efficient cooling systems—which all major firms are already adopting, suggesting the technology is ready but the incentive structure is not.
Is Big Tech’s Sustainability Reporting Enough?
No. Meta’s 51% increase in water usage from 2020 to 2024 demonstrates that closed-loop cooling systems and efficiency commitments are not keeping pace with growth. Google’s site-specific disclosures are a step ahead, but they still exclude third-party operated data centers. Microsoft and Amazon lag further behind. Without standardized, mandatory reporting of total consumption—including indirect water via electricity—investors and communities remain in the dark about the true cost of AI infrastructure.
Will Amazon, Microsoft, and Google comply with shareholder demands?
The companies have not yet committed to the level of transparency investors are demanding. Amazon’s statement about increasingly disclosing site-specific data suggests incremental progress rather than comprehensive disclosure. Google’s silence on requests for facility visits or interviews at its Uruguay site indicates resistance to external scrutiny. Microsoft’s continued refusal to break out data center water separately from total operations water suggests the company views granular disclosure as competitively sensitive. Shareholder pressure at 2026 annual meetings will determine whether that calculus changes.
How does data center water consumption compare across companies?
Google leads in transparency, publishing owned and leased facility data, though it excludes third-party operations. Meta reports owned sites only, which understates total consumption. Amazon reports per-power-unit metrics rather than totals, making year-over-year comparison difficult. Microsoft publishes only aggregate operational water figures. None of the three major cloud providers disclose U.S.-specific data center consumption separately, making it impossible for investors or communities to assess regional impact.
The data center water consumption crisis is not a future problem—it is happening now, and the companies building AI infrastructure are betting that opacity will outlast public attention. Shareholder resolutions are the first serious mechanism to force transparency. Whether they succeed will determine whether the AI boom is sustainable or simply shifted the environmental cost from Silicon Valley’s balance sheet to drought-stricken communities with no say in the matter.
Edited by the All Things Geek team.
Source: Tom's Hardware
