AI Boom Strains US Power Grids, Water Supplies, Driving Investor Scrutiny

The rapid expansion of Artificial Intelligence demands significant resources, straining power grids and depleting water supplies across the United States[4][13][15][22]. Data centers in the U.S. currently consume 4.4% of the nation's energy, projected to increase by 130% by 2030[4][22].

This resource strain introduces regulatory uncertainty for developers, raises concerns for local communities, and prompts investors to re-evaluate the financial risks of AI infrastructure[1][3][5].

What the records show

AI's rapid growth demands vast computational resources. Computing power usage increased tenfold between 2018 and 2022[4]. The process where AI models analyze new data and produce outputs, known as inference, now accounts for 80-90% of AI computing resources[4]. Data centers in the U.S. consume 4.4% of the nation's energy, a figure projected to rise by 130% by 2030[4][22]. AI workloads are expected to contribute almost half of the net increase in global data center consumption between 2024 and 2030[22]. This surge in energy demand occurs while over 60% of global electricity still comes from fossil fuels, risking higher greenhouse gas emissions[4].

Beyond electricity, AI infrastructure requires massive amounts of water for cooling data centers[12][13][21]. A medium-sized data center can consume up to 300,000 gallons of water daily, equivalent to the daily water usage of approximately 1,000 households[16]. Large facilities can consume as much as 5 million gallons daily[17]. Approximately 40% of U.S. data centers are located in areas experiencing high or extreme water stress[15]. These facilities often use open-loop evaporative cooling systems that continuously withdraw water, potentially lowering groundwater levels and straining municipal supplies, especially during droughts[14][16][18]. The discharge from these systems, known as blowdown, can elevate salinity and introduce chemical additives into local waterways, impacting water quality[14].

The AI boom also intensifies demand for critical minerals like gallium, germanium, indium, palladium, and tantalum, which are essential for high-performance chips and cooling systems[7][8][9][10][11]. China controls significant portions of the global supply for some of these minerals, such as 98% of primary gallium production and 60% of germanium refining, creating supply chain vulnerabilities[8]. The United Nations University Institute for Water, Environment and Health (UNU-INWEH) reported that AI-related water consumption could equal the basic annual domestic needs of 1.3 billion people by 2030[13]. It also projects up to 2.5 million tonnes of e-waste annually from AI infrastructure by 2030[13].

The timeline

Between 2018 and 2022, AI model training computational power usage increased tenfold[4]. In the first half of 2026, Diligent Market Intelligence identified 14 AI-related shareholder proposals to face a vote at U.S.-based companies, with nearly half raising concerns about data-center resource use[3]. By 2030, AI-related water consumption could equal the basic annual domestic needs of 1.3 billion people, and its land footprint could exceed 14,500 square kilometers, according to a UNU-INWEH report[13]. Also by 2030, the International Energy Agency (IEA) estimates data center energy demand in the U.S. will increase by 130%[22].

What changed and for whom

Communities are increasingly opposing data center projects due to concerns about noise, air pollution, environmental impacts, and quality of life[1]. This opposition influences local zoning and permitting decisions and prompts greater scrutiny from state lawmakers[1]. State policymakers are responding with legislation addressing permitting, electricity affordability, and water consumption for data centers[1]. For institutional investors, these pressures translate into financial risks, including project delays, rising infrastructure costs, and regulatory uncertainty[1][5]. Public pension funds are beginning to reconsider their exposure to data center investments[1]. Four of the five largest tech companies faced shareholder proposals during the recent proxy season, pushing for more accountability for the environmental and social risks of AI infrastructure[3]. Google and Amazon's net-zero climate goals are being challenged by their investments in artificial intelligence[19]. Google conceded that combating the energy and water-hungry needs of AI created a dilemma[19]. Amazon stated, "Though our emissions increased in 2025, we remain steadfast in our commitment to sustainability"[19].

What the subject has said publicly

Sam Altman, CEO of OpenAI, conceded that future AI research and development will consume more power than anticipated[22]. Microsoft "remains committed to making that possible—combining advancing capability with infrastructure" and is focused on "efficiency at scale"[20]. Microsoft's research finds that optimized AI inference consumes less energy than previously reported, with a median usage of 0.31 Wh per query[20]. The company states it uses carefully designed and specialized models and "smarter AI serving" to improve power utilization[20]. Microsoft has also transitioned its AI training workloads to a single pool managed by a machine learning technology called Project Forge to address utilization challenges[20]. Microsoft implements liquid cooling for IT hardware to significantly reduce the electricity required for thermal management compared to air-cooled servers[20].

What we could not confirm

We could not identify the specific locations of the 40% of U.S. data centers operating in water-stressed areas[15]. The exact number of public pension funds re-evaluating data center investments was not specified[1]. The precise extent to which community opposition has delayed or halted data center projects was not detailed[1].