Amazon this week published figures showing its global data center fleet consumed just 0.12 liters of water per kilowatt-hour in 2025 - a rate the company says is more than seven times more efficient than the industry average of 0.84 liters per kilowatt-hour, and a 52% improvement on its own 2021 baseline of 0.25 liters per kilowatt-hour.

The announcement, published June 11, 2026 on Amazon's news site and attributed to Amazon, frames the numbers against a backdrop of surging AI infrastructure demand and growing community concern over data center resource use. For the marketing and advertising technology industry, the disclosure matters because the same infrastructure that processes AI-powered ad targeting, auction bidding, and campaign measurement is responsible for the water and energy consumption being measured here.

How Amazon cools its servers - and when water enters the picture

Understanding the efficiency figure requires understanding the cooling architecture behind it. According to Amazon, roughly 90% of the time its data centers use what the company calls free air cooling - outside air is pulled in, passed over server rows to absorb heat, and expelled back outside. No water is involved. The approach works well in most climates for most of the year, and the operating principle is deliberately straightforward.

"It's kind of like in your house," said Joern Tinnemeyer, a data center engineering leader at Amazon. "It's a nice summer morning. It's not that hot out. I'm gonna open up my windows rather than turn the air conditioner on, and just let the breeze pull through."

The remaining 10% of operating hours - the hottest days in the warmest regions - is where evaporative cooling takes over. In this process, water is sprayed onto an absorbent medium and hot incoming air passes through it. As the water evaporates, it draws heat from the air, lowering temperatures by five to ten degrees Celsius before the air reaches the servers. Amazon water specialist Beau Schilz describes the medium as "a sophisticated, giant sponge."

"It's like sweating," Schilz explains. "The evaporative process pulls the heat off of your body so you don't overheat."

The alternative - mechanical chillers that operate like large-scale air conditioners - consumes significantly more energy. According to Amazon, chillers typically require 25% to 35% more electricity than evaporative cooling during peak hours. That extra demand arrives at precisely the moment when residential and commercial electricity consumption is highest, putting additional strain on already-stressed power grids. Amazon concluded that using water during the hottest days of summer is preferable to drawing heavily on grid capacity at peak load.

Running hotter to use less water

The 52% efficiency gain since 2021 did not come from a single technical leap. According to Amazon, it came from a sustained multi-year programme of raising the temperature thresholds at which its servers are designed to operate. If servers can tolerate higher ambient temperatures before cooling systems activate, the number of hours per year requiring evaporative cooling shrinks.

After several years of iteration and testing, Amazon now engages water cooling only when ambient temperatures exceed approximately 85 degrees Fahrenheit (roughly 29 degrees Celsius). The threshold means that in most temperate climates, the evaporative system sits idle for the vast majority of the year.

"This is how we innovate at Amazon," Tinnemeyer said. "We set an ambitious target that benefits our customers, iterate relentlessly, and validate with data - in this case, proving we could cut water use in half without any impact on performance."

The validation process was rigorous. Amazon analyzed thousands of hours of operational data across multiple data center campuses, tracking server failure rates as temperature thresholds were progressively raised.

"The failure rate didn't increase," Schilz said.

The practical impact of that finding was significant. "Our engineers looked at two identical data centers on the same campus and were able to use about 50% less water in one of them that was running with higher temperatures," Schilz said. In Northern Virginia - Amazon's largest region by IT load - water use dropped 42% year-over-year even as computing demand continued to grow.

The water positive goal and where things stand

Efficiency alone does not satisfy Amazon's stated ambitions. According to Amazon, the company has committed to becoming water positive by 2030, meaning that for every gallon consumed in data center operations, more than a gallon will be returned to local communities. As of the June 2026 announcement, Amazon says it is 75% of the way to that goal. In 2025, the company returned three gallons of water to communities for every four it consumed.

The total water withdrawal figure provides scale. In 2025, Amazon withdrew approximately 2.5 billion gallons across its entire global data center footprint for the full year. Amazon offers a comparison: Americans use roughly 3.3 trillion gallons annually for lawn and garden irrigation alone, according to the US Environmental Protection Agency. That means residential landscaping consumes more than 1,300 times the water that Amazon's data centers do.

At the sites Amazon directly owns and operates, total water withdrawal fell 2% from 2024 to 2025, even as the number of buildings in its global portfolio continued to grow.

Three strategies underpin the water-positive programme. The first two focus on reducing freshwater demand from Amazon's own facilities. The third involves investing in external water replenishment projects.

On the replenishment side, Amazon has announced more than 50 water projects globally, which it says are expected to return more than 5.8 billion gallons annually once fully operational - equivalent to filling approximately 8,800 Olympic swimming pools. Individual projects illustrate the geographic range. In Hermiston, Oregon, Amazon is funding a scheme to capture Columbia River water during peak winter flows and store it in an aquifer for use during drier months. In Pina del Ebro in northeastern Spain, in partnership with environmental consultancy Mediodes, a pipeline moves agricultural runoff to a poplar grove, reducing the need to draw irrigation water from the Ebro River. Near Guadalajara in Mexico, a watershed restoration project is designed to improve groundwater absorption and reduce pollution flowing into the Santiago River basin.

Reclaimed water - the infrastructure play

The third strand of Amazon's water strategy involves sourcing cooling water from reclaimed supplies - treated wastewater from municipal facilities - rather than from drinking water sources. Amazon says it currently operates 26 facilities running on 100% reclaimed water, and has 130 more under contract globally. The company claims this is more reclaimed water usage than any other cloud provider.

In three locations - Mississippi, Hong Kong, and Indonesia - Amazon has gone further, helping to establish reclaimed water programmes where none previously existed, funding the required infrastructure and assisting with permitting processes.

"We're not just using reclaimed water," said Usman Khan, a water specialist at Amazon for over seven years. "We're helping communities develop these programs from the ground up."

Schilz, who spent years working in water utilities before joining Amazon, framed the priority in terms of geographic targeting. "We're particularly focused on areas where there's water scarcity," he said. "Amazon wants to partner with communities to ensure our water stewardship creates local benefits that they want to see."

The question that the metric does not answer

Amazon's 0.12 L/kWh figure is a meaningful engineering benchmark. But it is a measure of energy consumed, not of computational output - and that gap matters when the primary driver of data center expansion is AI workloads where tokens generated, queries processed, and ad auctions run are the actual units of value.

Converting liters per kilowatt-hour into liters per token requires knowing how many tokens a kilowatt-hour of compute produces. That number is not fixed. It varies by model size, hardware generation, batch size, server utilization, and whether the workload is training or inference. A small open-weight model running at full utilization on modern hardware might produce 5 million or more tokens per kilowatt-hour. A large frontier model running inference on older chips might produce a fraction of that. No cloud provider currently publishes a verified tokens-per-kilowatt-hour figure.

Using a midpoint assumption of roughly 1 million inference tokens per kilowatt-hour - a reasonable but unverified estimate based on published chip benchmarks - Amazon's 0.12 L/kWh rate implies approximately 0.00012 milliliters of water per token, or around 0.06 milliliters per 500-token query. But that figure is a stacked estimate built on assumptions, and the uncertainty spans at least one order of magnitude in either direction. Any per-token water figure published without those caveats should be treated with caution.

Industry context and competing approaches

Amazon's 0.12 L/kWh figure sits well below the 0.84 L/kWh industry average it cites. The document notes figures from Microsoft, Google, and Meta as comparative data points, though it does not reproduce specific competitor numbers.

The broader industry is navigating competing pressures. Meta's data center in Newton County, Georgia, consumed approximately 10% of the county's total daily water supply, according to a 2025 investigation, with nearby residents reporting well water failures and pressure drops. Meta has separately announced billions in AI infrastructure investment, with its own reported water usage effectiveness figure of 0.20 L/kWh - better than the industry average but higher than Amazon's stated 2025 figure.

Google has taken a different approach in some new projects, opting for air-only cooling that eliminates evaporative water use entirely. Google's Meitner Energy Center in Gray and Roberts Counties, Texas, announced June 4, 2026, uses advanced air-cooling technology, limiting water consumption to small-scale domestic uses. Google's new data center in Horndal, Sweden, which broke ground June 2, 2026, is also air-cooled, reflecting a deliberate choice to limit water draw in a region where the topic has generated community concern. Air cooling, however, typically requires 10% more energy than evaporative approaches, creating a trade-off that each operator resolves differently depending on local climate, grid mix, and available water sources.

As PPC Land has documented in its coverage of data center economics, the sector accounts for less than 0.5% of global industrial water use in aggregate. The local impact, though, can be disproportionate when facilities are concentrated in specific water basins.

Why this matters for the advertising and marketing technology industry

Amazon's data center infrastructure is not incidental to the advertising industry - it is load-bearing. Amazon Web Services generated $30.9 billion in revenue and $10.2 billion in operating income in the third quarter of 2025, accounting for roughly 70% of Amazon's total profit. The same compute capacity that Amazon is measuring in liters per kilowatt-hour also runs Amazon Marketing Cloud, the DSP, the sponsored ads auction, and the AI inference behind Alexa for Shopping and advertising automation tools.

Amazon plans to spend approximately $200 billion on data center and AI infrastructure in 2026 alone. That investment is the foundation on which the company's advertising business - which crossed $70 billion in trailing twelve-month revenue as of Q1 2026 - depends. The water efficiency figures published today are therefore also, indirectly, a disclosure about the resource cost of running programmatic advertising at scale.

For marketers and agencies, the sustainability profile of the infrastructure they rely on is becoming a procurement and reporting consideration. Scope 3 emissions accounting - which captures the indirect environmental impact of purchased services - requires knowing what the upstream compute cost of running campaigns actually is. Water use, while not yet a standard line in Scope 3 frameworks, is increasingly part of the conversation as corporate sustainability teams scrutinize cloud vendor disclosures.

Amazon's 0.12 L/kWh figure, if independently verified, would give procurement teams a more defensible baseline than the industry average to use in supplier assessments. The company has not yet published site-level water consumption figures of the kind that Google began disclosing in late 2022 - a step that would allow more granular comparison across specific regions where data centers and water stress intersect.

Timeline

Summary

Who: Amazon, through its data center engineering and water specialist teams including Joern Tinnemeyer and Beau Schilz, with comparisons drawn against the broader data center industry including Meta and Google.

What: Amazon announced that its global data center operations achieved a water efficiency rate of 0.12 liters per kilowatt-hour in 2025 - a 52% improvement on its 2021 baseline and more than seven times better than the industry average of 0.84 L/kWh. The company also disclosed that it now operates 26 facilities on 100% reclaimed water, has 130 more under contract, has announced over 50 water replenishment projects expected to return 5.8 billion gallons annually, and is 75% of the way to its 2030 water-positive commitment.

When: The announcement was published on June 11, 2026, drawing on 2025 full-year operational data.

Where: Amazon's global data center footprint, with specific reference to Northern Virginia (its largest region by IT load, where water use fell 42% year-over-year), Hermiston Oregon, Pina del Ebro in Spain, Guadalajara in Mexico, and facilities in Mississippi, Hong Kong, and Indonesia where reclaimed water programmes were built from scratch.

Why: As AI-driven demand accelerates data center expansion globally, water use has become a material regulatory, community, and reputational issue for cloud operators. Amazon's disclosure positions the company against an industry average while establishing a quantified baseline for its 2030 water-positive pledge - one that carries direct relevance to the advertising and marketing technology ecosystem that runs on its infrastructure.