They hum in the darkness on the edges of cities, behind security fences and razor wire, their rows of blinking servers invisible to the billions of people whose digital lives they power.

Every Google search, every Netflix stream, every AI chatbot conversation, every Bitcoin transaction — it all flows through these vast, climate-controlled warehouses of computing power.

And they are consuming the Earth’s energy at a rate that is beginning to alarm scientists, regulators, and entire national power grids.

The data centre is the defining infrastructure of the 21st century. It is also becoming one of its defining environmental crises.

This is the story of how the world’s hunger for digital services — and particularly for artificial intelligence — is reshaping not just the global economy, but the global climate.

It is a story measured in terawatt-hours and megatonnes of carbon. And it is accelerating faster than almost anyone predicted.

The Environmental Impact of Data Centres

A PROBLEM HIDING IN PLAIN SIGHT

Cast your mind back to 2008. That year, global data centres consumed approximately 153.5 terawatt-hours (TWh) of electricity — a figure that seemed enormous at the time.

It was not yet enough to cause serious alarm. The internet was growing, yes, but energy efficiency improvements in server hardware were largely keeping pace with rising demand. For a decade, experts reassured themselves that the digital economy’s footprint was manageable.

Those reassurances are now obsolete.

By 2022, global data centre electricity consumption had reached somewhere between 240 and 340 TWh per year, according to the International Energy Agency (IEA).

By 2024, that figure had climbed to approximately 415 TWh — around 1.5% of all electricity used on Earth. And by 2025, with AI demand accelerating beyond all prior forecasts, consumption was approaching 500 TWh and rising steeply.

The IEA’s central scenario now projects data centre electricity consumption reaching 945 TWh by 2030 — more than six times what it was in 2008, in less than a generation.

To put that in human terms: the world’s data centres collectively consume more electricity than most nations. They already out-consume countries like France, and they are growing faster than almost any other sector of the global economy.

The carbon cost is equally stark. In 2020, global data centres were responsible for approximately 330 million metric tonnes of CO2-equivalent — around 0.9% of energy-related greenhouse gas emissions, according to the IEA.

In the United States alone, data centres emitted roughly 105 million metric tonnes of carbon in 2023, a figure comparable to the annual exhaust of more than 20 million cars. And this is before the full weight of the AI revolution has been felt.

HOW DATA CENTRES POLLUTE

To understand why data centres have such an enormous environmental footprint, you have to understand what they actually are: not just buildings full of computers, but extraordinarily complex, energy-hungry ecosystems that must run without interruption, 24 hours a day, 365 days a year.

Inside a typical data centre, thousands of servers and IT devices run continuously, drawing electricity to process and store data.

That electricity, in most parts of the world, is generated primarily from fossil fuels — coal, natural gas, and oil — and every kilowatt-hour burned releases carbon dioxide and other greenhouse gases into the atmosphere.

According to the Environmental Protection Agency, data centres account for 40% of the energy used by U.S. tech companies. Nationally, the 5,426 data centres operating in the United States as of March 2025 collectively consumed about 56% of their power from fossil fuel sources.

But the servers themselves are only part of the problem. Running at full capacity, thousands of processors generate tremendous heat.

That heat has to go somewhere. Cooling systems — the air conditioners, chillers, fans, and increasingly the liquid cooling rigs that ring every server farm — are responsible for more than 40% of a data centre’s total electricity usage.

An average Google data centre consumes approximately 450,000 gallons of water every single day just to keep its systems from overheating.

Across the industry, water consumption is staggering: U.S. data centres alone are estimated to have an indirect water footprint approaching 800 billion litres annually, linked to electricity generation.

Global data centre cooling energy demand was projected to reach 145 TWh by 2025, according to the Uptime Institute.

The U.S. Department of Energy, recognising the scale of the problem, has invested $40 million in research and development of liquid cooling technologies — next-generation systems that target heat precisely where it is generated, eliminating the need for wasteful evaporative cooling and capable of saving up to 90% of annual water usage.

Then there is a dimension of data centre pollution that most people never think about: backup power. Because these facilities must operate without interruption, they rely on massive banks of diesel generators as insurance against power outages.

Smaller facilities might house a handful. But the vast hyperscale centres being built to power artificial intelligence can require dozens — sometimes hundreds.

Quantum Loophole’s Aligned Data Centers, for instance, proposed installing 168 diesel generators capable of delivering 504 megawatts of power. The generators range in size from 1.5 MW to over 3 MW each, and they do not simply sit idle.

They are tested monthly, each test releasing significant quantities of particulate matter, nitrogen oxides, sulfur dioxide, and carbon dioxide into the surrounding air — pollutants that degrade local air quality and pose real health risks to nearby communities.

The people who live in the shadow of a data centre campus often know nothing of the invisible toll being extracted from their air and water. That is beginning to change.

THE AI EXPLOSION: A MULTIPLIER OF EVERYTHING

If traditional data centre growth was a slow-burning concern, the arrival of large-scale artificial intelligence has poured accelerant on the fire.

The numbers are almost difficult to absorb. Training a single large AI model can emit as much carbon as five cars over their entire operational lifetimes — a fact first quantified by researchers at the University of Massachusetts Amherst and one that has haunted the industry ever since.

A generative AI training cluster can consume seven to eight times more energy than a conventional computing workload.

AI workloads, according to Intel’s data, consume three to four times more energy than traditional IT tasks. The power density of server racks — the intensity of energy drawn from a given space — has increased by 30% in just the last five years as AI hardware has grown more powerful and more power-hungry.

The scale of infrastructure being built to support this demand is without historical precedent. In 2024, global data centre infrastructure spending reached approximately $290 billion.

Technology giants — Alphabet, Microsoft, Amazon, and Meta — together invested close to $200 billion of that total. In 2025, that spending was expected to climb by over 40%.

Combined investments from Microsoft, Amazon, Google, Meta, and Apple alone were forecast to exceed $450 billion in 2025. McKinsey projects a $7 trillion race to scale data centre infrastructure through 2030.

The energy implications are being tracked in real time by every major grid operator on Earth. The IEA estimated that approximately 2,700 data centres in the United States alone accounted for over 4% of the nation’s electricity use, a figure projected to rise to 6% by 2026 primarily due to AI adoption.

Goldman Sachs has forecast that data centres will account for 8% of U.S. energy usage by 2030 — more than double their current share.

At the extreme end of projections, data centres’ total U.S. electricity demand could reach up to 130 GW by 2030, equivalent to roughly 12% of total annual national demand.

In Ireland — a small Atlantic nation that has quietly become Europe’s de facto technology hub — data centres already consume around 21% of national electricity, a figure the IEA estimates could climb to 32% by 2026.

In the U.S. state of Virginia, the world’s most significant data centre market, these facilities already account for 26% of all electricity consumed. In Dublin, the situation has become so acute that it consumes an extraordinary 79% of the capital’s electricity supply, according to analysis by the Oeko-Institut.

And as emissions from most economic sectors begin to decline, data centres stand out as one of the few industries where emissions are set to grow.

The IEA estimates data centre carbon output will reach 1% of global CO2 emissions by 2030 in its central scenario — and 1.4% if growth accelerates. In a world straining to hold global warming below 1.5 degrees Celsius, that trajectory matters enormously.

AMERICA’S DATA CENTRE BOOM: A NATION REWIRED

The United States has no rival when it comes to data centre dominance. The country hosts nearly half of all data centres worldwide, with major concentrations in Virginia (643 facilities), Texas (395), and California (319).

The number has grown with breathtaking speed: in 2018, the country hosted approximately 1,000 data centres consuming about 11 GW of electricity, representing 1.9% of annual U.S. electricity consumption.

By March 2025, that number had ballooned to 5,426 facilities. The number of data centres more than doubled in three years from 2018 to 2021, then more than doubled again in the four years that followed.

The electricity that feeds this infrastructure — some 176 TWh consumed in 2023 alone — was drawn approximately 56% from fossil fuels.

And the carbon cost has been paid quietly, in warming skies and degraded air around the communities that host these facilities.

In 2018, U.S. data centres emitted around 68 million metric tonnes of CO2 — roughly the equivalent output of 14 million automobiles. By 2023, that had risen to approximately 105 million metric tonnes.

The grid is straining in ways that are starting to frighten regulators. In 2024, 60 data centres in northern Virginia simultaneously disconnected from the grid due to a tripped safety mechanism.

The resulting surge of excess electricity nearly caused a massive regional blackout. Network operators implemented emergency countermeasures just in time.

The near-miss exposed a fragility in American energy infrastructure that few had anticipated — and one that is growing more acute with every new server farm that comes online.

Building new fossil-fuel power plants to meet the demand would compound the problem enormously. The U.S. Environmental Protection Agency, the Department of Energy, and congressional members across party lines have all begun demanding better emissions disclosure from the industry.

AUSTRALIA: THE PACIFIC’S RISING DATA CENTRE FRONTIER

On the other side of the world, Australia is experiencing its own data centre reckoning — and the numbers are accelerating rapidly.

In the 2024-25 financial year, Australian data centres consumed an estimated 3.9 TWh of electricity, representing approximately 2% of total national grid-supplied consumption.

That may sound modest, but the trajectory is alarming. Under the Australian Energy Market Operator’s (AEMO) Step Change scenario, data centre consumption in Australia is forecast to grow at an average annual rate of 25.1%, reaching 12 TWh by 2030 and 34.5 TWh by 2050.

The Clean Energy Finance Corporation (CEFC) has warned that without significant new renewable generation and storage, data centres could account for up to 11% of Australia’s total electricity use by 2035 — up from around 1-2% today.

Hyperscale data centres — the vast AI-ready campuses operated by the likes of AWS, Google, and Microsoft — are already reshaping Australia’s energy landscape.

Amazon Web Services has announced an A$20 billion investment in Australian data centre infrastructure between 2025 and 2029, including plans for three new solar farms in Victoria and Queensland.

Between 2.2 GW and 3.2 GW of data centre capacity is expected to be operational in Australia by 2035, up from just 0.3 GW in 2024-25. That represents up to $135 billion in investment.

A 1 GW hyperscale data centre is reportedly being planned for Sydney — if built, it would be one of the largest in the world, consuming nearly as much power as half of Victoria’s Loy Yang A coal plant.

The carbon figures are already moving in the wrong direction.

New data from Australia’s federal Clean Energy Regulator shows that while overall direct emissions fell in 2024-25, indirect emissions from data centres rose from approximately 1.79 million tonnes of CO2-equivalent in 2023-24 to around 2.0 million tonnes in 2024-25 — a near-16% increase in a single year.

The surge was largely driven by major operators including Amazon, AirTrunk, and CDC, each of which recorded annual scope two emissions increases above 20%. Since 2020-21, these three companies alone have reported combined scope two increases of more than 100%.

The economic consequences for ordinary Australians are also real. CEFC modelling warns that without new renewable energy and storage investment, data centre growth could increase wholesale electricity prices by 26% in New South Wales and 23% in Victoria by 2035 — costs that would ultimately be borne by households.

In Ireland, where the data centre buildout preceded Australia’s by a decade, facilities were responsible for 88% of all increased electricity demand from 2015 to 2024. Australia is watching that cautionary tale unfold in real time.

“When big tech is given the green light for another energy-draining data centre, it’s not their billionaire backers who end up paying the price,” one industry critic told Australian media. “That’s paid by locals with spiking power and water bills.”

CRYPTO: THE ENERGY HUNGER NOBODY WANTS TO TALK ABOUT

If the data centre industry’s environmental impact is underappreciated, cryptocurrency mining’s footprint is even more poorly understood by the general public — and it is immense.

Bitcoin mining alone consumed an estimated 173 TWh of electricity in 2025, according to multiple analyses — representing approximately 0.5% of the world’s entire electricity consumption, comparable to the annual energy use of countries like Poland or Argentina.

A single Bitcoin transaction now requires roughly 1,335 kWh of electricity — enough to power an average American household for over 45 days.

Compare that to a Visa transaction, which consumes between one and five watt-hours. Bitcoin uses hundreds of thousands of times more energy per transaction than conventional digital payments.

The carbon cost is equally sobering. Bitcoin mining’s annual carbon footprint is estimated at approximately 98 million metric tonnes of CO2 — a figure comparable to the entire annual emissions of Qatar.

Cryptocurrency mining overall — including altcoins and other proof-of-work networks — consumed approximately 119.7 TWh of electricity in 2023, resulting in carbon emissions of around 90.6 million tonnes of CO2-equivalent, according to peer-reviewed research published in the journal Sustainability in April 2025.

Without significant intervention, those figures are projected to increase sixfold by 2030.

The International Monetary Fund has found that crypto mining could generate 0.7% of global carbon dioxide emissions by 2027.

The IMF calculates that a direct tax of just $0.047 per kilowatt-hour would bring the crypto mining industry’s emissions in line with global climate goals — and raise $5.2 billion in annual government revenue while cutting annual emissions by 100 million tonnes.

Where does all this mining happen? The United States now leads global Bitcoin mining with approximately 37.8% of global hashrate following China’s 2021 mining ban — a ban that dispersed operations to countries with varying electricity standards and environmental oversight.

Coal remains a significant energy source for Bitcoin mining globally, with fossil fuels still supplying 48% of the electricity used by the network in 2025, though that figure is gradually improving as miners increasingly seek out cheap renewable power.

Water consumption is another hidden cost. Bitcoin mining operations consume approximately 2,772 gigaliters of water annually — equivalent to Switzerland’s total annual water usage.

THE HUMAN GEOGRAPHY OF POLLUTION

The environmental damage caused by data centres is not distributed evenly. It falls disproportionately on the communities closest to these facilities — communities that often have little political power and limited awareness of what is being built in their backyards.

Diesel generators at major data centre campuses — sometimes dozens of them, each ranging from 1.5 to over 3 MW — are tested monthly, releasing particulate matter, nitrogen oxides, and sulfur dioxide into the local air.

These are not abstract pollutants. Particulate matter is linked to respiratory disease, cardiovascular problems, and premature death.

Nitrogen oxides contribute to smog and ground-level ozone. Sulfur dioxide causes acid rain. The communities that bear these health burdens are rarely the communities whose executives and engineers enjoy the services these centres power.

Water is another front line of impact. A single large data centre can consume millions of gallons of water per day through evaporative cooling systems, drawing on local aquifers and municipal supplies in ways that can strain water-scarce regions.

CONCERN GROWS — AND ACCOUNTABILITY BEGINS

As the evidence mounts, a clear shift is underway in how businesses, regulators, and civil society are responding to the environmental impact of data centres.

Major technology companies, including Google, Microsoft, and Meta, have pledged carbon neutrality or net-zero targets.

The reality, however, has not matched the rhetoric: recent reports show carbon emissions at these companies have actually increased as AI-driven data centre expansion accelerates.

Google’s own sustainability reports confirm this, even as the company invests in renewable energy and more efficient hardware. Microsoft, Amazon, and others are in a similar position: their climate pledges are being outpaced by their own infrastructure growth.

In Europe, data centre operators and industry associations launched the Climate Neutral Data Centre Pact in January 2021, pledging to make European data centres climate-neutral by 2030.

The European Union has set intermediate targets for power usage effectiveness and carbon-free energy by 2025. Progress is uneven.

In Australia, the Albanese government has announced plans for “national data centre principles” to align the industry’s explosive growth with national sustainability interests.

Regulatory pressure is intensifying everywhere. In the United States, members of Congress have pushed for mandatory disclosure regimes for data centre emissions and energy use.

New York state now requires environmental assessments for proof-of-work cryptocurrency mining. Over 40 U.S. states introduced or considered cryptocurrency mining legislation between 2024 and 2025.

Businesses, meanwhile, are under growing pressure not just to reduce emissions but to understand and account for them accurately.

As concern grows over the environmental impact of data centres, Fair Supply’s carbon emissions accounting software is emerging as part of the response for businesses under pressure to better understand and report their emissions

The pressure is real and growing. Investors, consumers, regulators, and communities are all demanding greater transparency.

The days when a technology company could build a data centre campus and describe it simply as “cloud infrastructure” — without accounting for its fuel mix, its water consumption, its diesel generators, or its carbon footprint — are drawing to a close.

🌍 World’s Most Power-Hungry Data Centres (2025–2026)

By Individual Facility Power Capacity

By Company / Hyperscaler (Total Fleet Consumption)

🥇 Amazon (AWS) The largest electricity consumer of the four major hyperscalers, Amazon consumed roughly 30+ TWh annually across its global data centre operations.

🥈 Microsoft Microsoft consumed approximately 13 TWh in 2021 and has grown dramatically since, with tens of billions now invested annually in new AI-focused data centres.

🥉 Google (Alphabet) Google consumed approximately 18.3 TWh in 2021, while maintaining an industry-leading PUE of 1.09 as of 2025 — meaning it wastes far less energy per unit of compute than competitors.

🏅 Meta Meta’s capital expenditure reached $39 billion in 2024, with spending potentially reaching $100 billion in 2026 — driven by projects like Prometheus, its first gigawatt-scale data centre in New Albany, Ohio, and Hyperion, a planned 5 GW campus in Louisiana covering roughly the footprint of Manhattan.

By Geographic Market (Regional Clusters)

Northern Virginia is the world’s single largest data centre market, hosting approximately 4,000 MW of capacity and nearly 300 individual facilities, many of which are AWS.

Beijing and London rank second and third globally, with the three cities combined accounting for over 5,400 MW — more than all other markets in the world combined.

THE ROAD FORWARD: WHAT CAN ACTUALLY BE DONE

The scale of the problem is daunting. But solutions exist — some implementable immediately, others requiring longer-term structural change.

The most powerful lever is the energy source. Renewable energy — solar, wind, hydropower, geothermal — produces electricity without direct carbon emissions.

Many data centre operators are pursuing it aggressively. Amazon is the world’s largest corporate purchaser of renewable energy, and its Australian expansion includes new solar farms.

Google, Apple, and Microsoft have all achieved or committed to 100% renewable energy matching for their data centre operations.

70% of Australian data centre energy demand is already linked to renewable sources through various purchasing mechanisms.

But renewable energy matching is not the same as real-time renewable power. Feeding a data centre’s constant, baseload demand with intermittent solar and wind requires massive investment in storage and grid infrastructure.

Meeting real-time demand with clean energy is, as energy experts acknowledge, a far more complex challenge than buying renewable energy certificates after the fact.

Advanced cooling technologies represent another major opportunity. The U.S. Department of Energy’s $40 million investment in liquid cooling research targets the systems responsible for more than 40% of data centre electricity use.

Modern liquid cooling systems that target heat at the source — rather than chilling entire rooms of air — can save up to 90% of annual water usage while dramatically cutting electricity demand.

Microsoft has already launched a new data centre design that consumes zero water for cooling, capable of avoiding more than 125 million litres of water per year per facility.

Emissions capture and control systems offer a solution that can be deployed right now, without waiting for grid decarbonisation.

Those systems retrofit onto diesel generators — filtering and retaining the exhaust of particulate matter, nitrogen oxides, and other pollutants every time a generator runs, whether during an outage or a monthly test. For communities living near large data centres, this technology represents immediate, tangible relief.

Smart scheduling is emerging as another tool. AI training runs — the most energy-intensive workloads — can, in principle, be shifted to times when renewable energy is most abundant on the grid.

Google has already adopted this approach. Australia’s energy experts have called for regulations requiring data centres to inform power companies in advance of large-scale AI training runs that cause dramatic energy spikes.

And then there is the question of efficiency itself. The most energy-efficient leading data centres already use around 84% less energy than the industry average, according to the World Economic Forum.

Spreading best practices — better hardware, hot and cold aisle containment, higher server temperature tolerances, custom AI chips designed for performance per watt — could significantly bend the consumption curve.

Google’s Ironwood tensor processing unit, its latest AI chip generation, is claimed to be 30 times more energy-efficient than its first publicly available TPU. Progress is real. It is simply not fast enough.

THE RECKONING ARRIVES

We stand at an extraordinary juncture.

The digital economy, which was once described as a solution to the physical world’s environmental problems — dematerialising products, virtualising services, reducing the need to travel — is now itself one of the fastest-growing sources of carbon emissions on the planet.

The AI revolution, whose advocates promise it will accelerate solutions to climate change, is simultaneously demanding an energy infrastructure that threatens to undermine the very climate targets those solutions are meant to protect.

The global data centre market consumed 153.5 TWh in 2008. It will consume roughly 945 TWh by 2030. In less than a generation, the digital world’s appetite for power has grown more than sixfold — and the growth is accelerating, not decelerating.

The servers will keep humming. The AI models will keep training. The Bitcoin network will keep mining.

The question is whether the world will choose to power all of that on coal and gas and diesel generators, or whether it will build the renewable energy infrastructure fast enough, and smart enough, to meet a demand that has caught almost everyone off guard.