AI’s build-out promises many economic benefits, not least higher productivity. But these long-term gains are being overshadowed by the near-term costs of the expansion.
Because data centres – the physical foundations of AI – consume vast amounts of water, electricity and land, their construction is facing fierce resistance from communities which are worried about having to pick up the tab.
In the US, home to three quarters of global data centre construction by capacity, what began as a scattering of local protests has escalated into a national data centre backlash that cuts across political divides. The US is home to 17 GW of 23.1 GW of capacity under construction globally. https://about.bnef.com/insights/data-centers/ai-data-center-build-advances-at-full-speed-five-things-to-know/
More than 20 data centre projects were cancelled in the first quarter of 2026, a record quarterly pace, after 20 withdrawals for the whole of last year which included Microsoft’s USD 3.3 billion facility in Wisconsin and QTS’s USD 30 billion development in Indiana.
In a related development, a dozen states have introduced restrictions on new facilities, while others are considering legislation to halt the data centre “invasion” altogether.https://www.interconnectedcapital.com/research/data-center-moratoriums
As a result, only about half of the capacity scheduled for the next two years is expected to come online.https://www.goldmansachs.com/insights/articles/us-data-center-power-demand-projected-to-double-by-2027
Such a rapid shift in public sentiment is forcing developers and operators to redesign facilities and adopt technologies to reduce the strain on local resources.
This involves the deployment of microgrids to ease pressure on local utilities, pairing renewables with battery energy storage to smooth peaks in demand, and installing water recycling systems to reduce competition with residential and agricultural users.
For investors, the backlash creates opportunities beyond tech, as demand grows for companies providing solutions that can help data centre operators shrink their environmental footprint to win local support and secure approvals.
"500 ml per response"
Public resistance to industrial projects is not new. Many may recall Erin Brockovich, the environmental activist portrayed by Julia Roberts in the 2000 film, who led a legal battle over water contamination in Hinkley, California, winning a USD 333 million settlement against a utility giant – then the largest payout of its kind.
Three decades on, Brockovich is back, this time taking aim at AI data centres.
She has launched a crowdsourced map that helps communities track their rapid, resource-intensive expansion in real time. Brockovich is not against data centres as such – her concern is about the lack of transparency and about environmental strain surrounding how these facilities are approved and built.
Industry experts say public opposition linked to data centres’ growing resource footprint is now the single biggest project risk in the US. This makes it a commercial priority for operators to allocate some of the USD 3.3 trillion they have set aside for AI capital projects through 2029 to smarter designs and technologies to reduce environmental impact and improve efficiency.BNEF
Efficiency measures can reduce local resource strain, key to winning local support, while helping operators maintain growth.
Reducing water consumption is particularly important. A typical data centre uses 300,000 gallons of water each day to keep servers and other equipment cool, equivalent to the consumption of some 1,000 households. For large facilities, this can rise to 5 million gallons each day, enough to supply a town of up to 50,000 residents.https://www.brookings.edu/articles/ai-data-centers-and-water/
Worryingly, demand for water is forecast to rise more than eight-fold in the coming years as more facilities come online, each requiring greater computing power to keep up with AI demand.https://www.ceres.org/resources/reports/drained-by-data-the-cumulative-impact-of-data-centers-on-regional-water-stressOne study estimates that a model such as ChatGPT effectively “drinks” a 500 ml bottle of water for every 10-50 medium-length responses.https://arxiv.org/pdf/2304.03271
Making matters worse, the industry’s search for cheap renewable power is pulling data centres towards sunny but dry regions such as Arizona and Texas. That creates a difficult water/carbon trade-off: sites with plentiful renewable power may help reduce carbon emissions, but they can also intensify pressure on already stretched local water supplies. More than 60% of over 800 planned US sites are due to be built in drought-hit areas, where many existing centres are already located.https://www.theguardian.com/us-news/2026/jun/08/datacenter-ai-drought-water
To mitigate these risks, data centres are turning to more efficient cooling technologies that are already available. For example, liquid cooling allows them to manage higher heat loads and operate around the clock. Rack manifold systems distribute the coolant through a stainless-steel central spine with parallel ports, ensuring even cooling.
Other operators are using treated wastewater and rainwater. Amazon, for example, has expanded the use of recycled water at its data centres to 120 US locations from 24, which it says will preserve 530 million gallons of water each year.
It plans to make its data centre operations “water positive” by 2030, having already improved its water use efficiency by 40% to 0.15 litres per kilowatt-hour of IT load – half the industry average of 0.36 litres per kilowatt-hour.https://sustainability.aboutamazon.com/stories/just-back-from-weftec-why-does-ai-use-water and https://escholarship.org/uc/item/32d6m0d1
Other efficiency measures, including smart data centre designs that constantly adapt to workload with AI chip accelerators and low-power processors, can reduce local resource strain. This is key to winning local support, while helping operators maintain growth.
Powering AI without overloading the grid
Electricity consumption is another problem. Globally, data centres’ electricity consumption has grown 12% per annum over the past five years, and it is expected to double to 945 terawatt-hours (TWh) by 2030.https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai
As AI accelerates the roll-out of high-performance servers, strains on power systems are becoming more visible.
In the US, which has seen AI power usage triple to 100 TWh in 2023 from 30 TWh in 2014,United States Data Center Energy Usage Report: 2025 Update, Lawrence Berkeley National Laboratory electricity costs have risen 42% over the past five years, outpacing the 29% rise in the overall consumer price index.https://www.brookings.edu/articles/confronting-and-addressing-rising-energy-bills-linked-to-data-centers/
Aware that energy costs are a political flashpoint, President Donald Trump introduced the Ratepayer Protection Pledge in March 2026, aiming to protect consumers from electricity price rises by requiring tech firms to shoulder the cost of powering AI data centres.
Half the US states have either introduced, passed or enacted special rate classes for large energy users, with some requiring data centres to fund infrastructure improvements and demonstrate benefits to consumers.
At the same time, hyperscalers are taking matters into their own hands by investing in power infrastructure upgrades and smart energy solutions. This includes the building microgrids – a localised grid network that generates, stores and distributes energy – high-voltage substations and bridging and fast power solutions such as backup generators and fuel cells.
Renewables and battery energy storage systems are also likely to see increased investment. Data centre operators are expected to ramp up green power purchases – hyperscalers already account for more than half of clean power purchase agreements last year, and collectively procured a record 28 GW of clean power in 2025.
Despite the growing outcry, the current US pipeline points to a much larger data centre footprint, more than double today’s if fully built.https://cleanview.co/data-centers/us
Greater scrutiny will not stop the boom, but it should change capital allocation plans, forcing operators to spend faster on advanced water treatment and cooling, smarter design, sustainable energy and grid infrastructure.
