Greenpeace report How Clean is Your Cloud? I saw mentioned in 3T magazine news is actually quite interesting reading. This year’s report provides a look at the energy choices some of the largest and fastest growing IT companies. The report analyzes the 14 IT companies and the electricity supply chain in more than 80 data center cases.
The report contains also lots of interesting background information on both IT and telecom energy consumption. I recommend checking it out. Here are some points picked from How Clean is Your Cloud? report:
Facebook, Amazon, Apple, Microsoft, Google, and Yahoo – these global brands and a host of other IT companies are rapidly and fundamentally transforming the way in which we work, communicate, watch movies or TV, listen to music, and share pictures through “the cloud.”
The growth and scale of investment in the cloud is truly mind-blowing, with estimates of a 50-fold increase in the amount of digital information by 2020 and nearly half a trillion in investment in the coming year, all to create and feed our desire for ubiquitous access to infinite information from our computers, phones and other mobile devices, instantly.
The engine that drives the cloud is the data center. Data centers are the factories of the 21st century information age, containing thousands of computers that store and manage our rapidly growing collection of data for consumption at a moment’s notice. Given the energy-intensive nature of maintaining the cloud, access to significant amounts of electricity is a key factor in decisions about where to build these data centers. Industry leaders estimate nearly $450bn US dollars is being spent annually on new data center space.
Since electricity plays a critical role in the cost structure of companies that use the cloud, there have been dramatic strides made in improving the energy efficiency design of the facilities and the thousands of computers that go inside. However, despite significant improvements in efficiency, the exponential growth in cloud computing far outstrips these energy savings.
How much energy is required to power the ever-expanding online world? What percentage of global greenhouse gas (GHG) emissions is attributable to the IT sector? Answers to these questions are very difficult to obtain with any degree of precision, partially due to the sector’s explosive growth, a wide range of devices and energy sources, and rapidly changing technology and business models. The estimates of the IT sector’s carbon footprint performed to date have varied widely in their methodology and scope. One of the most recognized estimates of the IT sector’s footprint was conducted as part of the 2008 SMART 2020 study, which established that the sector is responsible for 2% of global GHG emissions.
The combined electricity demand of the internet/cloud (data centers and telecommunications network) globally in 2007 was approximately 623bn kWh (if the cloud were a country, it would have the fifth largest electricity demand in the world). Based on current projections, the demand for electricity will more than triple to 1,973bn kWh (an amount greater than combined total demand of France, Germany, Canada and Brazil).
The report indicates that, due to the economic downturn and continued energy efficiency and performance improvements, global energy demand from data centers from 2005-2010 increased by 56%. Estimates of data center electricity demand come in at 31GW globally, with an increase of 19% in 2012 alone. At the same time global electricity consumption is otherwise essentially flat due to the global recession is still a staggering rate of growth.
Given the scale of predicted growth, the source of electricity must be factored into a meaningful definition of “green IT”. Energy efficiency alone will, at best, slow the growth of the sector’s footprint. The replacement of dirty sources of electricity with clean renewable sources is still the crucial missing link in the sector’s sustainability efforts according to the report.
The global telecoms sector is also growing rapidly. Rapid growth in use of smart phones and broadband mobile connections mean mobile data traffic in 2011 was eight times the size of the entire internet in 2000. It is estimated that global mobile data traffic grew 133% in 2011, with 597 petabytes of data sent by mobiles every month. In 2011, it is estimated that 6 billion people or 86.7% of the entire global population have mobile telephone subscriptions. By the end of 2012, the number of mobile connected devices is expected to exceed the global population. Electronic devices and the rapidly growing cloud that supports our demand for greater online access are clearly a significant force in driving global energy demand.
What about telecoms in the developing and newly industrialized countries? The report has some details from India (by the way it is expected that India will pass China to become the world’s largest mobile market in terms of subscriptions in 2012). Much of the growth in the Indian telecom sector is from India’s rural and semi-urban areas. By 2012, India is likely to have 200 million rural telecom connections at a penetration rate of 25%. Out of the existing 400,000 mobile towers, over 70% exist in rural and semi-urban areas where either grid-connected electricity is not available or the electricity supply is irregular. As a result, mobile towers and, increasingly, grid-connected towers in these areas rely on diesel generators to power their network operations. The consumption of diesel by the telecoms sector currently stands at a staggering 3bn liters annually, second only to the railways in India.
What is the case on other developing and newly industrialized countries? I don’t actually know.
NOTE: Please note that that many figures given on the report are just estimates based on quite little actual data, so they might be somewhat off the actual figures. Given the source of the report I would quess that if the figures are off, they are most probably off to direction so that the environmental effect looks bigger than it actually is.
942 Comments
Tomi Engdahl says:
https://www.facebook.com/share/1EnVXPAWfk/
Sähkönnälkää: Lahden DayOne datakeskuksen sähkön tuotantoon tarvittaisiin seitsemän Kalantia, Suomen suurinta aurinkopuistoa.
Yksi tuottaa sähköä 20 000 kotitalouden tarpeisiin: seitsemän 140 000 kotitalouden – tai yhden Lahden datakeskuksen.
Orimattilan suunniteltuun 800MW:n datakeskusalueeseen näitä Suomen suurimpia aurinkopaneelipuistoja tarvittaisiin 35 kappaletta, tai 700 000 kotitalouden tarpeen verran.
Uudenkaupungin Kalantiin valmistunut aurinkovoimala on Suomen suurin, ja se tuottaa sähköä jo valtakunnan verkkoon. Puisto havainnollistaa hyvin, kuinka paljon sähköä nykyaikaiset datakeskushankkeet vaativat toimiakseen ympäri vuorokauden.
Kalannin aurinkovoimalapuiston alta on poistettu 200 hehtaaria maa-ja metsätalousmaata 40 vuodeksi.
Kalannin puisto lukuina
Helenin omistama Kalannin aurinkovoimala kattaa noin 200 hehtaarin eli kahden neliökilometrin alueen 12 paneelikentässä. Voimalan nimellisteho on 206 megawattia, ja vuotuinen sähköntuotanto on Helenin mukaan yli 200 gigawattituntia.
Puistoon asennettujen paneelien tarkka lukumäärä on vaihdellut eri vaiheiden uutisoinnissa: alkuperäisissä suunnitelmissa puhuttiin 380 000 paneelista, kun taas tuoreimmat, valmiin puiston läpikäyneet raportit puhuvat noin 360 000–362 000 paneelista. Ero selittynee rakennusvaiheen tarkentumisella suunnittelusta toteutukseen.
Paljonko sähköä datakeskus tarvitsee?
Aurinkopuiston 200 gigawattitunnin vuosituotanto vastaa keskimäärin noin 22,8 megawatin jatkuvaa tehoa (200 000 MWh jaettuna vuoden 8 760 tunnilla). Jos tällä sähkömäärällä pyöritettäisiin datakeskusta, jonka energiatehokkuusluku (PUE) olisi 1,2, varsinaiselle IT-laitteistolle jäisi käytettäväksi noin 19 megawattia.
Laskelma on kuitenkin teoreettinen: aurinkopuisto ei tuota sähköä tasaisesti ympäri vuorokauden tai vuoden, vaan 206 megawatin huipputeho toteutuu vain kirkkaimpina hetkinä keskellä päivää kesällä. Yhden aurinkopuiston varaan rakennettu datakeskus tarvitsisi väistämättä tuekseen sähköverkon, energiavarastoja tai muuta tuotantoa tasaamaan vaihtelua.
Lahteen Kiveriön alueelle rakennettava DayOnen datakeskus on mitoitettu 128 megawatin IT-kuormalle täydessä laajuudessaan – tämä luku viittaa nimenomaan IT-laitteiston tehoon, ei koko laitoksen sähkönkulutukseen. Jos datakeskuksen PUE olisi esimerkiksi sama 1,2 kuin edellä, laitoksen todellinen kokonaissähkönkulutus olisi noin 154 megawattia ja vuosikulutus noin 1 350 gigawattituntia – tämä vastaisi noin seitsemän Kalannin kokoisen aurinkopuiston vuosituotantoa.
Orimattilan Pennalaan kaavaillun datakeskuksen arvioitu kapasiteetti on kaupungin omien kaavadokumenttien mukaan 800 megawattia täydessä laajuudessaan. Kaavamateriaalissa ei eritellä, viittaako luku IT-kuormaan vai koko laitoksen sähkönkulutukseen. Jos 800 megawattia otetaan sellaisenaan jatkuvana kokonaistehona, vuosikulutus olisi noin 7 000 gigawattituntia – noin 35-kertainen määrä Kalannin puiston tuotantoon verrattuna.
Tomi Engdahl says:
https://www.facebook.com/share/p/1EDF6GaXnr/
As the race for artificial intelligence intensifies, China is moving some high-performance computing infrastructure beneath the ocean to reduce the energy required for cooling.
Underwater data centers are already operating off Hainan and Shanghai. Their sealed server modules use seawater as a natural cooling source, reducing reliance on conventional refrigeration systems.
For the Shanghai project, developers have reported that cooling can account for around 40% to 50% of electricity consumption in conventional facilities, while the underwater design can reduce the cooling share to below 10%. The operational Shanghai facility has also been reported to reduce overall electricity consumption by 22.8% and eliminate freshwater use compared with a conventional land-based facility.
The technology could therefore reduce the environmental footprint associated with cooling, water consumption and land use. At the same time, large-scale deployment requires careful monitoring of potential environmental effects, including how discharged heat could affect surrounding marine ecosystems over time.
As underwater computing expands, its long-term sustainability will depend not only on energy efficiency but also on responsible design and continued environmental assessment.
Images are generated by AI and for demonstration purposes only.
Source: Xinhua. (2025). World’s first wind-powered commercial underwater data center project launched in Shanghai. Xinhua, n/a, n/a.
#technology #artificialintelligence #ai #datacenter #innovation #sustainability #china #fblifestyle
Tomi Engdahl says:
https://www.facebook.com/share/p/1DF71ejZYw/
The UN’s warning isn’t a headline. “Water insolvency” is what happens when demand beats nature’s ability to refill the tank. Aquifers built over centuries are being spent in decades. Rainfall is shifting. Farms, factories, and cities are pulling from the same shrinking account. The fix is boring and proven: manage better, waste less, recycle more, and irrigate smarter. Protecting freshwater isn’t optional. It’s the constraint everything else now sits inside.
Tomi Engdahl says:
https://www.facebook.com/share/p/19WoAR3KCv/
Data centers consume staggering amounts of water to cool their servers—water that’s then released as vapor into the atmosphere or drained as contaminated wastewater into local waterways. In drought-stricken regions, this means communities face higher utility costs, restricted water access, and depleted aquifers so that tech companies can store our photos and emails.
The technology exists to solve this. Closed-loop cooling systems, recycled wastewater integration, and air-cooling alternatives are all viable. But they cost more, and without regulation, companies choose the cheapest option—which is always someone else’s problem.
Here’s the tension: societies need digital infrastructure. But they also need clean drinking water, which no technology can replace. When those needs collide, which one should lose? The answer depends entirely on who holds the power to decide. Right now, it’s not the people whose wells run dry.
Source: MOST Policy Initiative, Data Center Water Use, 2026.
Tomi Engdahl says:
How about instead of melting Antarctica, build them in cold countries some distance away from cities in industrial area, and use their excess heat to keep the buildings warm without need to burn oil, gas and wood?
Power those data centers mainly from renevables and CO2 free power sources.
Would that be a good practical idea?
Tomi Engdahl says:
Randy Joyce Data centers primarily use clean municipal drinking water (potable water) for evaporative cooling systems, though many modern facilities increasingly transition to treated wastewater (recycled or gray water) or closed-loop liquid systems to conserve freshwater resources.Types of Water UsedFresh Potable Water: Standard municipal tap water is the historical and current default because it minimizes mineral buildup, scaling, and biological growth in cooling towers.Reclaimed / Recycled Wastewater: Treated sewage effluent or gray water is increasingly used by major operators like Amazon, Google, and Microsoft to reduce strain on local drinking water supplies.Dielectric / Specialized Liquids: Used in modern direct-to-chip or immersion cooling setups, these non-conductive fluids circulate in sealed, closed loops rather than evaporating.
Tomi Engdahl says:
Susi Art I checked your claim rather than dismissing it, and there’s an important distinction.
You’re correct that some data centers use potable municipal water for cooling. That’s well documented, and I agree that using drinking water for cooling deserves scrutiny especially where water is scarce.
But potable water isn’t required for data-center cooling, nor does every facility use it. Reclaimed water, non-potable sources, closed-loop cooling and other technologies are already being used.
I’m perfectly willing to change my position when the evidence supports it. But if we’re making claims this strongly, we should be able to verify the sources being cited.
Tomi Engdahl says:
While AI data centers dominate headline anxiety over local drinking supplies, agricultural crops like alfalfa consume vastly more water on a macro scale.
Data Centers (Direct Consumption)
National Total: ~0.14% to 0.2% of total U.S. water consumption. Over 40% of data centers are built in high water-stress regions. In specific watersheds (like Northern Virginia’s Potomac River Basin), data centers can account for around 3% of all local water used.
Alfalfa (Agricultural Irrigation)
National Total: ~6% to 8% of total U.S. freshwater withdrawals nationwide.
Western US / Colorado River Basin: ~20% to 32% of all water consumed across the American West.
The Regional Reality: “Exporting Water”
West Coast Export Share: ~30% to 38% of all alfalfa grown in states like California and Washington is shipped abroad.
Tomi Engdahl says:
The concern about automated surveillance networks—like Flock Safety cameras—is grounded in real, documented developments. Tens of thousands of Automated License Plate Readers (ALPRs) collect billions of vehicle scans monthly, creating nationwide, searchable databases used by police departments and local governments.
However, conflating that real-world surveillance system with the physical buildout of commercial AI data centers conflates two completely different technological and business ecosystems.
The hardware stack required for consumer AI products like ChatGPT or Gemini is completely different from what is needed to process and index camera footage.
Modern Flock cameras run computer vision models right on the device at the edge using small solar-powered processors, uploading plain text metadata (plate numbers, color, make) to standard cloud servers. Storing text logs of vehicle scans takes up a fraction of a percent of normal cloud storage—it does not require gigawatt-scale AI supercomputing clusters.
Who Owns the Infrastructure?
The multi-billion-dollar AI data centers dominating current headlines are built by public commercial companies (Microsoft, Amazon, Google) and specialized private cloud providers (CoreWeave, Lambda Labs).
Their financial viability depends on charging enterprise clients, developers, and consumers for cloud software, API tokens, and web services.
Flock Safety is an independent surveillance vendor selling hardware and software subscriptions to municipal police departments, private security, and neighborhood associations. It hosts its services on commercial cloud networks, but it doesn’t build or run 100-megawatt AI campuses.
Tomi Engdahl says:
Yes, water vapor is technically the single most abundant greenhouse gas in Earth’s atmosphere and accounts for the majority of the natural greenhouse effect. However, calling it “worse” than \text{CO}_2 misses a fundamental distinction in how atmospheric physics and climate feedback loops operate.
Water vapor does not drive climate change on its own; it acts as an amplifying feedback, whereas \text{CO}_2 acts as the primary driver.
Tomi Engdahl says:
water vapor is technically the single most abundant greenhouse gas in Earth’s atmosphere and accounts for the majority of the natural greenhouse effect. However, calling it “worse” than \text{CO}_2 misses a fundamental distinction in how atmospheric physics and climate feedback loops operate.
Water vapor does not drive climate change on its own; it acts as an amplifying feedback, whereas CO2 acts as the primary driver.
Water Vapor (\text{H}_2\text{O}): Stays in the atmosphere for roughly 9 to 10 days. If you pump extra water vapor into the air—whether from cooling towers, ocean evaporation, or rain—it simply condenses and falls back down as rain or snow within a week and a half.
Carbon Dioxide (\text{CO}_2): Remains in the atmosphere for 300 to 1,000+ years. Once emitted, it accumulates continuously, creating a persistent, long-term heat-trapping layer that does not fall out with the weather.
Tomi Engdahl says:
https://www.facebook.com/share/p/1EcM8PiZ2E/
AI may feel weightless and digital, but the infrastructure powering it has a very physical environmental cost.
Texas is rapidly becoming a major hub for AI data centers, and some developers are installing on-site gas turbines and diesel generators to meet enormous electricity demands. An investigation found that at least 38 Texas data centers have obtained permits commonly used for “minor” pollution sources—raising concerns about limited public notice and environmental review.
These generators can release carbon dioxide, nitrogen oxides and other pollutants. Nitrogen oxides contribute to smog and can aggravate asthma and other respiratory conditions, especially in nearby communities.
The wider climate footprint could also be substantial. Cornell University researchers estimate that rapid growth in U.S. AI servers could produce 24–44 million metric tons of carbon dioxide-equivalent emissions annually by 2030—comparable to adding roughly 5–10 million gasoline-powered cars to American roads. The same research projects annual water consumption of up to 1.125 billion cubic metres.
AI’s environmental impact is not inevitable. Cleaner electricity, more efficient computing, stronger emissions standards and transparent permitting could significantly reduce it. But without careful planning, the rush to expand AI could lock communities into years of additional fossil-fuel use, water pressure and local air pollution.
The digital future still requires power plants, water and land—and communities deserve to know the true cost.
Tomi Engdahl says:
“The digital future still requires power plants, water and land”
I can agree both power and land needs are always there.
The water need depends on how the data center is built. Some data centers and generating power for them uses lots of water. Some other data centers use almost no water at all.
For those data centers that use lots of water, tapping drinking water lines for server cooling in drought-prone or water-stressed regions creates direct competition between tech facilities and residential or agricultural needs (which are typically much bigger than data center use).
Freshwater Usage Virtually Zero (0) data centers:
Google’s Hamina facility consumes virtually no clean municipal freshwater for server cooling.
Because heat reuse and seawater from the Gulf of Finland handles 100% of server heat dissipation, municipal drinking water from the local utility (Haminan Vesi) is used strictly for non-server operations —typically under 20,000 to 40,000 liters (5,000–10,000 gallons) per day, comparable to a small local commercial office building or light industrial site.
Power Usage: ~200 MW to 300+ MW
Google has contracted over 255 MW of renewable energy (primarily Finnish onshore wind power) specifically to feed the Hamina grid connection.
Local utility Haminan Energia to capture waste heat from the servers and pump it into the city’s municipal district heating grid, supplying around 80% of Hamina’s residential heating demand.
Because Hamina bypasses municipal drinking water for heat rejection, its total daily impact on the town’s public water supply is roughly 99\% lower than a conventional evaporative data center of the same capacity.
Tomi Engdahl says:
The idea of building floating, solar-powered offshore data centers sounds like a simple fix, but mathematically and physically, the concept encounters major structural barriers.
While placing servers at sea solves land acquisition and drinking water issues, relying purely on ocean solar power falls short due to basic energy density and environmental constraints.
Solar Intermittency & Capacity Factor: Floating solar panels (floatovoltaics) have a capacity factor of ~15% to 20% due to night cycles, cloud cover, and seasonal sun angles. To get 100 MW of continuous power, you would need to build an array with a peak rating of at least 500 MW to 600 MW.
Massive Footprint: Producing 500 MW of solar power requires roughly 2 to 3 square miles (5 to 8 square km) of floating solar arrays.
Wave Damage: Covering square miles of ocean surface with fragile solar panels leaves them vulnerable to storm surges, saltwater corrosion, biofouling (barnacles/algae), and destruction from sea swells.
Solar panels generate zero power for 12 hours every night.
To keep a 100 MW data center running through the night, a floating platform would need to store 1,200 Megawatt-hours (MWh) of electricity every single day.
That would require one of the largest lithium-ion battery banks on Earth mounted on a floating barge.
Startups attempting offshore data centers avoid relying strictly on solar power.
Tomi Engdahl says:
https://www.facebook.com/share/1EVcqcvqAj/
Need context for just how bad data center emissions are? Here you go.
Just two planned UK data centers will produce more annual carbon emissions than ExxonMobil’s entire British operations.
Analysis by the tech justice non-profit Foxglove reveals that when fully operational, the proposed Wapseys Wood and Quest Park data centers will consume 1.3 gigawatts of power, producing over 4.5 million tonnes of carbon dioxide annually.
This surpasses ExxonMobil’s UK carbon output of 3.9 million tonnes in 2023. Facing long waitlists to connect to the national grid—which has 315 data centers currently queued, nearly doubling the UK’s peak winter energy demand—developers are planning to construct their own on-site gas-fired power plants to keep the servers running.
This massive energy demand highlights a growing conflict between the expansion of tech infrastructure and international net-zero commitments. In the United States, which consumes roughly 40% of the world’s data center electricity, the scale of this environmental footprint is even larger.
Research estimates that the 60 largest U.S. data centers currently under construction by tech giants like Amazon, Google, Meta, and Microsoft could release over 101 million tons of carbon dioxide annually when fully operational—equivalent to adding 24 million cars to American roads. As governments race to streamline permits, communities and climate advocates are increasingly questioning whether the societal and economic benefits of AI software justify the immense environmental toll.
source: Neill, P. (2026). Two planned datacentres will have higher UK carbon emissions than ExxonMobil, analysis finds. The Guardian.
Tomi Engdahl says:
https://www.facebook.com/share/p/1CMRPkXJeT/
U.S. AI data centers in just 7 states withdraw 3.4 trillion gallons of water annually, exceeding the water demand of all California cities combined.
The massive energy appetite of the artificial intelligence boom is hiding an environmental crisis. According to a landmark report by the sustainability nonprofit Ceres, data centers across seven key states—including Virginia, Texas, and California—indirectly withdraw roughly 3.4 trillion gallons of freshwater annually just to generate the electricity required to keep them running.
This staggering volume exceeds the average annual urban water demand of every city in California combined, exposing a massive, largely unmonitored strain on regional water resources.
While tech giants frequently boast of their local, on-site cooling efficiencies, they rarely disclose this indirect water footprint from the power plants feeding their facilities. Much of this power comes from water-intensive thermal or hydroelectric sources, with a significant portion located in regions already suffering from severe drought. As the AI arms race accelerates, experts warn that without strict transparency and a shift to water-efficient energy sources, this hidden water liability will directly compete with local communities for vital freshwater supplies.
source: Ceres. (2026). The Water Behind the Watts. Ceres.
Tomi Engdahl says:
Edward Velez modern data centers ARE A CLOSED LOOP SYSTEM. Any water that is used and not contained in that closed loop system is returned to the environment cleaner than it was to begin with. Direct to chip technology is nothing like old data center technology. Your not the only one who has worked with them or the department of environmental quality or that has a degree in environmental science. That’s exactly what.
Jeff Taylor
No, they are not 100% closed loop. They are hybrid systems. Most cooling systems like this are.
While the direct to chip and building cooling are closed loop. I would say 100% are cooled loop. What cools the closed loop cooling system is most often not.
I need you to imagine trying to cool the 149F to 185F temprature of a direct chip cooling system to something more acceptable when the outside temperatures are above 105F under the Texas sun. It doesn’t work. That’s what the water is used for. To cool down the external chillers using evaporative cooling.
Tomi Engdahl says:
Kevin Jones hyper-scale data centers dump up to one million gallons of waste water PER DAY, in what is know as a ” Blow Down”. This is when the ultra- toxic, heavy metal laden and the biocide chemicals get dumped. So they use millions of gallons of water, they evaporate 80% of that water. 20% is discharged and tracked/ dictated via CoC (Cycles of Concentration). They do NOT keep radiators full of water to reuse. They discharge daily at a rate up to 12 gallons per second, straight into storm drains usually. So take 1,000,000 gallons of discharge divided by 12 gallons per second… that is 24 hours. So they NEVER stop discharging toxic water.
Tomi Engdahl says:
there are currently some 4,000 open-loop systems in the USA alone. Of those, nearly 400 are hyper-scale. That’s a LOT of waste water.
Jim Rose
The facility counts in the claim align well with reality, but calling them “open-loop” systems that generate massive volumes of “waste water” is inaccurate. The real environmental footprint of these facilities is local water depletion through evaporation, not filling sewers with wastewater.
The Core Error: They Are Evaporative, Not Open-Loop
Calling these facilities “open-loop” systems and claiming they produce a “LOT of waste water” misrepresents how data center cooling mechanics actually operate:
Open-Loop Systems (Rare in Data Centers): An open-loop system continuously pulls water from a river or aquifer, runs it through the building once to absorb heat, and discharges almost all of it back down the drain as warm wastewater. Data centers rarely use true open-loop systems because of strict environmental discharge regulations.
Evaporative Cooling (The Actual Method Used): Most water-cooled data centers use evaporative cooling towers. Instead of dumping water down the drain, the system intentionally evaporates the water into the air to pull heat away from the servers.
The Real Environmental Issue: Lost Water, Not Waste Water
The primary environmental critique of traditional data center cooling is not “waste water,” but water consumption (evaporation):
Evaporated, Not Discharged: Up to 80% to 85% of the water drawn by an evaporative data center evaporates directly into the atmosphere. It is lost from the local municipal watershed rather than entering local sewer lines as wastewater.
Blowdown Water (Actual Wastewater): Only a small fraction (around 15% to 20%) is discharged as “blowdown” water—mineral-heavy water sent back to municipal treatment plants once it becomes too saturated to continue evaporating.
Tomi Engdahl says:
Water Recycling (Variable, but Rapidly Improving)
Whether a data center recycles its cooling water depends entirely on the design of the facility:
Closed-Loop Systems: Modern data centers increasingly use sealed liquid-to-chip or closed-loop air chilling systems. These systems recycle 100% of their internal cooling fluid endlessly in a closed cycle without losing water.
Reclaimed Water Use: Rather than drawing fresh drinking water, many operators contract with municipalities to use greywater (recycled municipal sewage effluent) for their evaporative cooling loops.
The Evaporation Exception: The one area where water is not recycled is in traditional evaporative cooling towers. Water is intentionally evaporated into the air to pull heat out of the building. That evaporated water leaves the local watershed rather than being recycled locally.
Tomi Engdahl says:
HR Woods
The one you are working in has smaller cooling needs that these AI data centers.
Let me explain.
Before I do, I’ll give you a little of my background. I have worked in around data centers for over 50 years. Yes, data centers have been around that long. Longer, as a matter of fact. I have built out data centers. Installed the communications equipment in data centers. Installed the processing power in data centers. I have even designed systems that produced that processing power. I will never call myself an expert. But I do have more knowledge than the average Facebook Keyboard warrior.
Back to your question. Many, if not most data centers use a closed loop cooling system. They is a Chiller somewhere on the property that feeds cool treated water or other coolant to an air handler in the computer room(s). That is an effective system. It works extremely well on those types of data centers. These new data centers generate much more heat. A lot more. Which requires a much larger and more efficient cooling system.
Depending on the size of the data center you work in. It to may still use some non-recycled water to cool the center. If see what looks like steam coming from the roof of the building or from some small structure at the back of the building. That is most likely water evaporating off the coils of a chiller. In a lot of cases a “closed loop” system is not 100% closed loop. The coolant may be. But what keeps the coolant cool is not.
Tomi Engdahl says:
go use this prompt if you want to burn some more tokens. “Of the hyperscalers operating in the US, how many are using an alternative to open loop cooling, followed by “of planned US hyperscaler projects, how many will use open loop cooling?”
I’d paste the answer here, but let’s burn a few more tokens.
Of the estimated 580 to 600 operational hyperscale data centers in the US, approximately 20% to 25% (roughly 115 to 150 facilities) currently use alternatives to traditional open-loop (evaporative) cooling.However, this metric is undergoing a massive split between legacy cloud facilities and newly constructed AI clusters”.
Tomi Engdahl says:
As an engineer, thus is the most ridiculous claim that even the most demented could imagine. Non- contact colling water systems have been in use fir a hundred years. Pricess industries do not want river contamination entering their systems and also do not want the massive costs, the amount of studies and permitting and treatment of “wastewater” that would result from open loop cooling. Your statements appear to imply that data centers open loop cool, contact corrode all of their capital equipment into oblivion, and deposit it into the waterways. Maybe this is a business plan in your world, but not in the real world that the rest of us live in.
Tomi Engdahl says:
The internet may feel weightless, but the infrastructure behind it is anything but.
Modern data centers need enormous amounts of electricity, land and, depending on their cooling systems and location, water. With AI driving demand higher, the IEA expects global data-center electricity consumption to roughly double by 2030, while electricity use by AI-focused data centers could triple.
Water is becoming a major concern.
Servers generate enormous amounts of heat, and many data centers rely on water-based cooling. A 2026 peer-reviewed study found that their impact can become especially significant when multiple facilities are concentrated in communities already facing water stress.
And the water footprint doesn’t stop at the data center.
Power plants that generate the electricity can also consume large amounts of water, meaning some of the water footprint is hidden upstream.
Then there’s land.
Large facilities can require hundreds of acres, new transmission infrastructure and supporting development. Communities across the U.S. are increasingly questioning projects because of concerns about farmland, water supplies and electricity demand.
None of this means data centers are inherently bad. They power cloud services, banking, communications, scientific research and the AI systems we’re increasingly using every day.
Better cooling, recycled water, cleaner electricity and smarter locations can reduce their environmental impact.
The question isn’t whether we need data centers.
We do.
The question is how we build them without putting the water, land and ecosystems communities depend on at unnecessary risk.
https://www.facebook.com/share/p/1F7vc4NHEh/
Tomi Engdahl says:
Across the United States, golf courses consume vastly more water directly than data centers. Golf courses also evaporate more water than data centers.
Those are small numbers compared to industry, energy and agriculture sectors.
Direct on-site data center water evaporation accounts for roughly 0.05% of agricultural water consumption in the United States. Even when adding the indirect water needed by power plants to run server farms, all U.S. data centers combined draw less than 1% of the water consumed by crop irrigation.
Tomi Engdahl says:
Heavy water consumption is an operational design choice, not an absolute physical law of computing.
Data centers can be—and increasingly are—engineered to operate with zero water loss.
A data center drawing millions of gallons of water is using legacy, low-cost evaporative engineering. The technology to run high-density AI clusters with zero water consumption is already deployed and operating at scale.
Tomi Engdahl says:
https://www.facebook.com/share/1LFob4ZT8L/
Jason Kelce wants your pee to cool AI data centers in bizarre new campaign
Liquid Death and Garage Beer have launched a bizarre satirical campaign starring former NFL player Jason Kelce, asking Americans to collect their urine and send it to AI data centers. The campaign mocks the huge amounts of water some data centers use for cooling.
Kelce appears in a humorous music video encouraging people to drink the brands’ products, collect their pee and mail it as supposed “data center coolant.” However, the advertisement clearly says not to actually send urine, making the request a publicity stunt rather than a real cooling program.
The campaign taps into growing public criticism of AI infrastructure over its water and electricity consumption. More than seven in ten Americans reportedly oppose new data centers being built in their communities, according to polling cited by multiple reports.
Rather than making a conventional environmental campaign, Liquid Death and Garage Beer are using shock humor to turn the data-center water debate into viral entertainment. The stunt highlights a genuine concern, while the companies’ deliberately outrageous solution is meant to grab attention rather than provide real cooling water.
Tomi Engdahl says:
Tekoäly lämmittää pian 70 000 helsinkiläiskotia
https://etn.fi/index.php/13-news/19247-tekoaely-laemmittaeae-pian-70-000-helsinkilaeiskotia
Tomi Engdahl says:
https://www.facebook.com/share/19NokhA2Y2/
Samsung Heavy Industries is reportedly developing a 50-MW floating data center designed to operate offshore instead of consuming valuable land onshore. The platform would use abundant seawater for cooling, potentially reducing dependence on freshwater cooling systems.
The concept could help address the growing demand for land, electricity and water created by AI data centers. Offshore facilities could also connect to coastal grids through underwater cables or potentially generate their own power onboard.
Samsung is working with partners including Capital Clean Energy Carriers, Lloyd’s Register and Supermicro to test the technology and marine operating conditions. Engineers must ensure AI servers can withstand vibration, humidity, salt exposure and other challenges that do not exist in conventional land-based facilities.
If successful, Samsung aims for a commercial launch around 2028, potentially opening a new path for large-scale AI infrastructure. Similar projects are emerging in Japan, China and the United States, suggesting that the future of data centers could increasingly move offshore.
Sources:
Samsung Heavy Industries
The Register
Tom’s Hardware
Tomi Engdahl says:
https://www.facebook.com/share/p/1CFGwpBWYq/
Erin Brockovich has turned her attention to one of the fastest-growing pieces of infrastructure in America: AI data centers. After receiving thousands of messages from concerned residents, she created a nationwide map tracking major facilities that are operating, under construction or proposed, alongside reports submitted by nearby communities.
What immediately raised concern was where many new facilities are going. A 2026 analysis found that roughly two-thirds of planned U.S. data centers were located in areas that had experienced drought during the previous year. That’s important because some of the largest facilities can require millions of gallons of water per day, depending heavily on their cooling technology and local climate.
The water footprint extends beyond cooling too. Berkeley Lab estimates cited in recent reporting put direct U.S. data-center water consumption at around 17 billion gallons in 2023, with substantially more water indirectly associated with producing the electricity they consumed.
That doesn’t mean every data center is draining its community’s water supply. Water use varies enormously depending on cooling systems, location and power source, and newer facilities can use closed-loop or low-water designs. The bigger issue Brockovich is raising is whether communities already facing water stress should have far greater transparency and a voice before enormous new facilities are approved.
Her campaign is increasingly about more than AI itself. It’s about what happens when the infrastructure behind our digital world begins competing with farms, wildlife and communities for physical resources in the real world.
Source: Erin Brockovich Data Center Reporting Project / Lawrence Berkeley National Laboratory / NOAA drought data.
Tomi Engdahl says:
Direct On-Site Emissions (What Leaves the Building)
Thermal Energy (Low-Grade Heat):
Nearly 100% of the electricity entering a server is converted directly into heat.
Water Vapor (Only in Evaporative Facilities):
If the facility uses evaporative cooling towers, it emits massive plumes of pure water vapor (steam) into the atmosphere.
In closed-loop or dry-cooled facilities, water vapor emissions are zero.
Intermittent Diesel/Gas Generator Exhaust:
Data centers maintain large backup diesel (or natural gas) generators to ensure uptime if the main grid fails.
During normal operations, these sit idle. However, routine monthly testing or power outages result in brief emissions of standard combustion byproducts: CO2, nitrogen oxides (NOx), and fine particulate matter.
Refrigerant Micro-Leaks (Scope 1 Fugitive Emissions):
Chiller systems use synthetic refrigerants (hydrofluorocarbons or newer low-GWP hydrofluoroolefins). Minor leaks over time emit small volumes of these gases, which have high global warming potential if not properly recovered during maintenance.
Logically, a data center is essentially a massive, high-tech electric space heater. It emits warm air, some noise from fans, and—unless connected to a clean grid—drives power plant emissions elsewhere on the electrical network.
Tomi Engdahl says:
Brockovich Data Center Reporting – U.S. AI Data Center Awareness …
Interactive map of major AI data centers across the United States — operational, under construction, permitted, proposed and cancelled.
https://www.brockovichdatacenter.com/
Tomi Engdahl says:
Brockovich Data Center Reporting – U.S. AI Data Center Awareness …
Interactive map of major AI data centers across the United States — operational, under construction, permitted, proposed and cancelled.
When the website officially launched, it went viral on social media and across news outlets, resulting in massive traffic spikes.
The ironies of an “anti-data-center website crashing due to excessive server traffic” became a widely shared meme across social tech forums.
Where is it hosted?
The site is hosted on standard commercial cloud web platforms using static site hosting and third-party mapping libraries:
Domain & Hosting: The standalone domain brockovichdatacenter.com relies on lightweight static web hosting (front-ended by modern Content Delivery Networks like Cloudflare to handle caching, mitigate traffic spikes, and prevent future downtime).
Tomi Engdahl says:
Fact: Communist China is funding opposition to data centers in America. The Communists want to dominate us and fools are falling for it. There are thousands of data centers that have been in operation for decades. They have never bothered anyone. They don’t have to make noise and they don’t consume water rather it’s recycled just like the water in your car radiator. It’s all lies for the weakminded. If you are lucky to get a data center in your county you can see your property taxes cut in half because the data center will be paying massive taxes. You will also see your electric bill lowered.
Tomi Engdahl says:
The chinese are flooding social media with anti US AI center propaganda. the newer AI centers use closed loop cooling. even the older AI centers use far less water than farming or golf courses. Also the newer AI centers , have their own on site electric power plant, off the grid. The chinese will do anything to win the AI war to dominate the world economy and intelligence. they plan to monitor everybody’s private activities. and to limit our military ability to defend our country and allies. Don’t be duped!
Tomi Engdahl says:
https://www.facebook.com/share/p/19UjKKzAJa/
Commerce Secretary Howard Lutnick has claimed that AI data centers “don’t use water” and dismissed concerns over their resource consumption, stating, “This is propaganda by our adversaries to try to slow us down.”
Speaking on CNBC, Lutnick added, “One of my favorite things is when people talk about data centers using water,” before arguing that “the No. 1 product in America that uses water is cattle.”
However, a 2024 Lawrence Berkeley National Laboratory report estimated U.S. data centers consumed 66 billion liters of water in 2023, with hyperscale facilities alone projected to consume between 60 billion and 124 billion liters annually by 2028 as AI infrastructure expands.
Tomi Engdahl says:
that’s 66B liters per year, or around 17B gallons per year. That’s out of 120 TRILLION gallons of water per year used in the US. So Data centers account for about 0.01% of US water usage, and could expand up to 0.02% by 2028.
When put in context, it is clear that water is a non-issue here
Tomi Engdahl says:
The claim’s conclusion is mostly right, with one key caveat:
Globally/Nationally: Water depletion from data centers is a complete non-issue compared to agriculture, power generation, and heavy manufacturing.
Locally: It remains a valid issue if developers build legacy evaporative facilities in water-stressed deserts instead of adopting zero-water closed-loop designs.
Tomi Engdahl says:
When including indirect water consumption (the water evaporated off-site at thermoelectric power plants to generate the grid electricity that powers data centers), the total footprint increases roughly 12-fold.
Tomi Engdahl says:
The Key Takeaway: Solvable via Decarbonization
The fact that 92% of a data center’s water footprint is indirect highlights a major operational reality:
PPA Procurement: When a tech company executes a Power Purchase Agreement (PPA) for 100% wind or solar energy, they eliminate both their carbon emissions and their off-site water footprint simultaneously, as wind and solar photovoltaics require virtually zero water during operation.
Closed-Loop + Green Grid: A data center equipped with closed-loop dry cooling that buys renewable power drops its total water footprint (direct + indirect) by over 98%, demonstrating how site engineering and grid selection dictate environmental impact.