The Good, the Bad, and the AI

All references cited in this article are part of the public record. The assertions and conclusions presented, unless otherwise noted, have not been legally contested.

Someone, somewhere, is typing it right now: AI data centres are wrecking the planet, draining the grid and drinking the rivers dry. They hit post. Within seconds, thousands of people have read it, liked it and shared it.

Every one of those steps ran through a data centre. The complaint was received by one, stored in several, ranked by an AI system running in one, and delivered to every reader from servers cached in data centres around the world. By the simple act of complaining online, the complainer became a customer. Everyone who complains about data centres on the Internet is, by default, a hypocrite.

That does not mean every concern about data centres is wrong. Some are serious, and this article gets to them. But the conversation starts from a misunderstanding of what data centres are, what they do, and how much of daily life already depends on them.

The Journey of a Complaint

Follow a single post. It leaves your phone, passes through a cell tower or your Internet provider, and lands on the platform's servers, in a data centre, where it is stored and copied to others so it is not lost if one fails.

Then it has to reach an audience. On Facebook, an AI system orders each person's feed by predicting what they will find most relevant, calculating a relevance score for about 500 candidate posts and ranking them.¹ That system runs in a data centre. So does the one that decides whether your post breaks the rules, and the one that suggests who else might want to see it.

When other people open it, most of what they see does not come from the platform's main servers at all. It comes from content delivery networks, systems of smaller data centres placed close to users that keep copies of pages, images and videos so they load quickly. Cloudflare alone operates in more than 330 cities in over 100 countries and says its network is within 50 milliseconds of about 95 per cent of the world's Internet users.² Every like, every share and every reply repeats the cycle.

An online complaint about data centres is not a protest against the system. It is the system, working exactly as designed.

Data Centres Were Here Long Before AI

The word "data centre" now conjures images of AI, but the buildings are not new, and most of what they do has nothing to do with chatbots. Your online banking and e-transfers run through them. So do your email, video calls, streaming services, online shopping, electronic health records, government services, the weather app on your phone, and the map that routes you around traffic. Hospitals, airlines, power utilities and emergency services all depend on them.

AI is a fast-growing share of the load. The International Energy Agency's analysis, as summarized by Carbon Brief, puts AI at roughly 5 to 15 per cent of data centre electricity use in recent years, possibly rising to 35 to 50 per cent by 2030.³ Even at the high end, that would leave half or more of the work doing what data centres have always done: Running the digital life nearly everyone already lives.

The Good

Scale is efficient. Power usage effectiveness, or PUE, measures how much energy a data centre uses beyond what goes into the computers themselves. Google reports a fleet-wide PUE of about 1.09, meaning roughly 9 per cent overhead for cooling and power. The industry average in Uptime Institute's 2024 survey was 1.56, or 56 per cent overhead.⁴ Large, modern facilities are far more efficient than the server closets and small data centres they replace.

A single AI query is smaller than people think. In August 2025, Google published the first large-scale measurement of AI energy use taken directly in production. It found the median text prompt to its Gemini assistant used 0.24 watt-hours of electricity, about as much as watching television for less than nine seconds, and 0.26 millilitres of water, about five drops. Over the previous 12 months, the energy used per median prompt had fallen by a factor of 33.⁵ That is one company's figure for one kind of request, and image, video and "reasoning" tasks use more. But it is a long way from the popular image of every question boiling a kettle.

AI is doing real science. The 2024 Nobel Prize in Chemistry went in part to Demis Hassabis and John Jumper of Google DeepMind for AlphaFold, an AI system that has predicted the structures of virtually all of the roughly 200 million proteins researchers have identified, a problem scientists had struggled with since the 1970s. It has been used by more than two million people in 190 countries.⁶ In February 2025, the European Centre for Medium-Range Weather Forecasts put its AI forecasting system into operation. It outperforms traditional physics-based models on many measures, including tropical cyclone tracks by up to 20 per cent, and uses about 1,000 times less energy to produce a forecast.⁷

The waste heat can warm homes. Data centres turn nearly all of their electricity into heat, and some cities have started using it. In Stockholm, heat from data centres is piped into the district heating network, warming about 10,000 apartments, with a goal of supplying 10 per cent of the city's heat. In Ireland, waste heat from an Amazon data centre heats buildings in Tallaght, and in Quebec, QScale's Lévis campus is designed to send surplus heat to neighbouring greenhouses.⁸

The Bad

The growth is real. The International Energy Agency estimates data centres used about 415 terawatt-hours of electricity in 2024, around 1.5 per cent of the world's total, and expects that to more than double to about 945 terawatt-hours by 2030. Electricity use by AI-focused data centres is expected to roughly triple.⁹ In the United States, data centres used 4.4 per cent of all electricity in 2023, and a U.S. Department of Energy-funded study projects 6.7 to 12 per cent by 2028.¹⁰

The impact is local, not global. Averages hide the places where data centres cluster. In Ireland, data centres used 23 per cent of all metered electricity in 2025, up from 5 per cent in 2015. Only households used more.¹¹ In Virginia, the world's largest data centre market, they account for about a quarter of the state's electricity.³ When one industry takes that large a share of a grid, everyone else feels it, in new power plants, new transmission lines and, often, in their bills.

Water is a genuine concern. U.S. data centres consumed about 17 billion gallons of water directly for cooling in 2023, and roughly 211 billion more indirectly through the power plants that supplied their electricity.¹⁰ In drought-prone regions, that competes with farms and towns. And companies rarely disclose, site by site, exactly how much they use.¹²

Power is only as clean as its source. Data centres produced about 0.5 per cent of global CO₂ emissions in 2024, according to Carbon Brief's analysis of IEA data.³ Where they run on hydro, wind, solar or nuclear, the footprint is small. Where they run on new gas plants, it is not.

Alberta's Bet

Canada has its own test case. Alberta has pitched itself as a destination for AI data centres, and demand has flooded in. The Alberta Electric System Operator capped new large loads at 1,200 megawatts to protect grid reliability, and by late 2025 had allocated all of it to two projects near Edmonton. Another 37 data centre projects are asking for a combined 19.4 gigawatts.¹³

The largest is Wonder Valley, near Grande Prairie, a proposed $70 billion project backed by Kevin O'Leary and billed as the world's largest AI data centre industrial park. It has sought approval for up to 9 gigawatts of power, planned largely around off-grid natural gas.¹³ Whether projects like that are good for Alberta depends on questions that deserve public answers: How much gas will be burned, how much water will be drawn, what happens to the emissions, and who pays if the promised jobs and revenue never arrive.

Putting It in Perspective

Data centres are growing fast, but they are not the main driver of rising electricity demand. Between 2024 and 2030, Carbon Brief's analysis of IEA projections shows data centres adding about 530 terawatt-hours of demand worldwide. Air conditioning is expected to add 651, electric vehicles 838, and industry 1,936. Data centres account for roughly 8 per cent of the growth, perhaps 12 per cent in a faster-growth scenario.³

That is the pattern the online outrage misses. Globally, data centres are a modest and manageable share of energy use. Locally, where they cluster on a small grid or in a dry region, they can be a serious burden. The problem is not that data centres exist. It is where they are built, what powers them, what cools them, and who carries the cost.

Where They Belong

There is a difference between putting a data centre on land that industry has already poisoned and putting one on farmland beside a rural aquifer. The outrage rarely makes that distinction. It should.

Consider Hamilton. For more than a century its steel mills baked coal into coke, and coke ovens release a mix of carcinogens including benzene and benzo(a)pyrene. The U.S. Environmental Protection Agency classifies coke oven emissions as a known human carcinogen.¹⁴ ArcelorMittal Dofasco was Ontario's single largest industrial greenhouse gas emitter in 2023, releasing 4.16 million tonnes, and its coke ovens emitted benzene and benzo(a)pyrene at levels far above provincial guidelines. Its site-specific exemption allowed benzene emissions 22 times the provincial standard.¹⁵ Dofasco shut its No. 3 coke plant on April 9, 2026.¹⁶

Now compare a large data centre drawing a steady 100 megawatts, about 876 gigawatt-hours a year. By my arithmetic, using the Canada Energy Regulator's grid figures, on Alberta's gas-heavy grid (470 grams of CO₂e per kilowatt-hour in 2022) that works out to roughly 410,000 tonnes a year, about a tenth of Dofasco. At the national average of 100 grams, it is roughly 88,000 tonnes, about a fiftieth. On Quebec's hydro grid, at 1.2 grams, it is closer to 1,000 tonnes, a few thousand times less.¹⁷ And a data centre has no coke ovens. It does not emit benzene or benzo(a)pyrene. Its main local air pollution comes from diesel backup generators when they are tested or run during outages, which is a real concern, but not one in the same class as a coke battery.¹⁸

On most Canadian grids, then, a data centre is somewhere between ten and a hundred times less carbon-intensive than a mill like Dofasco, and far cleaner than that on a hydro grid. Scale still matters. If Wonder Valley ever drew its full 9 gigawatts around the clock at a gas-fired intensity like Alberta's, it would emit in the range of 37 million tonnes a year, nearly nine times Dofasco. The type of building is only half the question. The other half is how big it is and what powers it.

Industrial land stays industrial. Where coke was made, homes are unlikely ever to stand. Cleanup standards for residential land are far stricter than for industrial land, and bringing a heavily contaminated site up to residential standards can mean digging out and hauling away soil and treating groundwater at enormous cost.¹⁹ In practice, the realistic choice for most former heavy-industry land is another industrial use, or nothing. Those sites often come with exactly what a data centre needs: Heavy-duty grid connections, water infrastructure, rail and road access, and industrial zoning.²⁰ That is why the former Homer City coal plant, once Pennsylvania's largest, is being redeveloped as a data centre campus, and why a developer has proposed a data centre with battery storage on the site of Scotland's Ravenscraig steelworks, which closed in 1992.²¹ In Hamilton itself, Slate Asset Management, which owns the roughly 800-acre former Stelco lands now branded Steelport, has proposed an AI data centre there, and is working with the city-owned Hamilton Community Enterprises on piping the waste heat into downtown buildings through district energy.²²

Rural land is a different story. Building on farmland or near private wells puts a new industrial user into a place that was never designed for one. In Newton County, Georgia, residents living beside a new Meta data centre reported their well water slowing to a trickle after construction began. Meta has disputed that its project was responsible, but the county, facing a surge of data centre development, is on track for a water deficit by 2030.²³ Replacing a coke plant with a data centre is an environmental improvement. Replacing a field and a clean aquifer with one is not.

Not in My Backyard

The data centre fight follows a familiar pattern: Everyone wants what the building provides, and nobody wants the building. It is the same argument that has dogged wind farms for decades. People want clean power, just not a turbine on their horizon.

But wind turbines are the clearest example of why things belong where they belong. They need strong, steady, unobstructed wind, which is exactly what cities do not have. Buildings break the wind into turbulent eddies that confuse a turbine and cut its output. Turbines mounted on or around buildings have missed their expected output by as much as 90 per cent, and a badly placed rooftop turbine can produce almost nothing even in a strong wind.²⁴ Nobody would put a turbine among the towers of a downtown core.

Safety points the same way. A turbine blade can fail, and ice can build up on blades and be thrown in winter. That is why turbines are set back from homes. Ontario, for example, requires industrial wind turbines to stand at least 550 metres from the nearest home, and more for larger groups.²⁵ In a city, a failure would send debris into streets, cars and windows. In open farmland, with proper setbacks, there is room for the risk to land harmlessly. Wind farms belong in rural areas because that is where the wind is, where the space is, and where a failure hurts no one. That is not a compromise. It is simply the right place.

The same logic applies to data centres, in the other direction. Turbines belong on open land. Heavy industrial users belong on industrial land, near the grid, near water that can be spared, and ideally on sites that industry has already used. Put each where it fits, and much of the "not in my backyard" fight disappears.

Alberta has managed to get this nearly backwards. In August 2023, the province paused approvals for new renewable energy projects for seven months. When the pause ended, it barred wind development within 35 kilometres of designated "pristine viewscapes," covering a large stretch of the foothills west of Calgary, and banned renewable projects on the province's best farmland unless developers could show that farming could continue alongside them.²⁶ Protecting prime farmland has a real rationale. But the same government that made it harder to put wind turbines on windy rural land, where they belong, is welcoming a gas-fired data centre park that could draw up to 9 gigawatts in the countryside south of Grande Prairie.¹³ That is not keeping industry out of the backyard. It is choosing which industry gets in.

What the Wind Replaces

A wind farm on a rural ridge is not just a turbine on the horizon. Every megawatt-hour it produces is a megawatt-hour that a fossil fuel plant does not have to make. In Alberta, University of Alberta researchers traced what wind actually displaced over a decade of real market data, as it grew from about 6 to 12 per cent of the province's electricity supply. In 2018, coal made up 84 per cent of the generation next in line to run when the wind was blowing. By 2023, as carbon pricing drove coal off the grid, natural gas made up 84 per cent. Between 2013 and 2023, Alberta's wind farms avoided about 39 million tonnes of emissions, roughly a full year of the province's entire electricity-sector output. In 2023 alone, wind supplied 12 per cent of the power but displaced nearly 17 per cent of the grid's emissions, because what it replaced was the dirtiest generation on the margin.²⁷ Today, in most of Canada, the choice on a windy day is between wind and burning gas, or coal.

Canada has made real progress on coal, but not everywhere:

Ontario was first. Its last coal-fired station, in Thunder Bay, burned its final coal in April 2014, making Ontario the first jurisdiction in North America to eliminate coal power. The province calls it the single largest greenhouse gas reduction action on the continent, equal to taking up to seven million cars off the road.²⁸

Manitoba phased out coal in 2018.²⁹ Alberta, which as recently as 2016 got close to half its electricity from coal, finished its phase-out in June 2024, largely by converting coal plants to natural gas.²⁷

Nova Scotia and New Brunswick still burn coal for power, with plans to stop by 2030, replacing it with wind, imports and other sources.²⁹

Saskatchewan is going the other way. Coal still makes up about 28 per cent of SaskPower's generation, and the province has committed $900 million to refurbish its coal plants. Federal rules require unabated coal power to end by December 31, 2029. Saskatchewan's minister responsible has said the province "will not comply," calling coal "a pillar of our electrical generation" until nuclear power arrives in the mid-2030s or later.³⁰

The territories do not burn coal for electricity, but Nunavut faces its own problem: Almost all of its power comes from diesel fuel, shipped in each summer and stored for the year, in 25 separate community power plants.³¹

That is the real context for a wind farm in the countryside. It is not a scar on someone's view. It is less coal and gas burned somewhere on the grid, and it makes every building that draws from that grid, including a data centre, cleaner as a result. Restrict the wind while welcoming gas-powered data centres, and you make both problems worse at once.

The Honest Complaint

There is a better way to complain, and it does not require logging off. Ask where a proposed data centre will be built: On land industry has already used, or on farmland and fresh groundwater? Ask whether the local grid and water supply can handle it. Ask whether it will run on clean power or new fossil fuel generation. Ask whether its waste heat will warm homes or be dumped into the air. Ask who pays for the new transmission lines. Ask companies to publish their energy and water use site by site, and ask governments to make that a condition of approval.

Those are questions worth posting. Just remember, when you hit send, that a data centre is carrying your question to everyone who reads it, and that the honest goal is not a world without them. It is a world where they are built in the right places, powered the right way, and held to account like any other industry.