Environmental Impacts of Data Centres – A SAGE Review (June 2026)
The Questions We Should Be Asking:
1. Electricity Generation: Is there a long-term plan for providing energy to generate the electricity required by data-centres? This might include renewable energy, natural gas (fracking), or nuclear power.
2. Electricity Rates: Will all data centres be expected to be isolated from Alberta’s grid? If not, how will consumers be protected from increases in the price of electricity?
3. Emissions: How does the Government of Alberta propose to meet (realistically) our net-zero emission targets for 2050 considering the massive energy demand from data centres? If the use of renewable energy was mandated for data centres, emissions would be reduced by 90%.
4. Land Use: The Government of Alberta has limited the use of Class 1 and 2 agricultural land for renewable energy projects. Will this same policy be applied to data centres? The Wonder Valley project is said to have 26,000 acres set aside for industrial use.
5 Water: Will the Government of Alberta restrict water abstractions from basins closed to new licenses? Will wastewater contaminants be monitored and controlled to protect aquatic ecosystems?
6. Noise: Will the Government of Alberta evaluate the siting of data centres to protect wildlife from disruption as a result of vibration and noise pollution? Will human populations be similarly protected?
7. Policy: Will the Government of Alberta engage in social or environmental impact assessments, public consultation, or regulatory oversight? Should we be bypassing regulatory review to fast-track the construction AI data centres?
8. Social Impacts: How will Alberta manage job loss? How will we manage educational outcomes? Do we always need to use AI for data inquiries or problem solving (or should it be only used where and when it has proven useful)? Should we ask how AI works in a substantive democracy (who has the power; how do we know AI outputs are trustworthy)? How do we manage biases perpetuated by AI? Should we tax AI data centres to pay for their environmental and social impacts?
Some Context:
The Government of Alberta has stated that the province is to “become a global leader in AI data center operations”, offering “advantages such as power generating capabilities, abundant natural resources, a cold climate, low corporate taxes, and reduced red tape.”i
The Government of Alberta wishes to leverage its natural gas reserves, reduce the need to expand pipeline capacity out of the province, and take advantage of the cooler northern climate. Already, there are roughly 38 data centres being considered in Albertaii, with Synapse Data Centre Inc. currently planning the largest in Canada near Olds.
To date, Canada has five ‘hyperscale’ data centres, with 96 more being plannediii. (Hyperscale denotes facilities requiring more than 100 MW of power). Alberta will host 90% of these projects with power requirements estimated to exceed 12 GW. To put this in perspective, this is more than the electricity normally generated in Alberta for all other usesiv.
There are a number of emerging issues with the rapid development of large data centres. These include: energy consumption (and associated emissions of greenhouse gases); water consumption (and pollution); changes in land-use; and regulatory retreat from public consultation and environmental impact assessment.
Energy Consumption:
[Primer: Power is the energy used in a period of time and is given the notation of gigawatts [GW], which is a billion joules of energy per second. Energy is presented in gigajoules [GJ], and represents the heat energy derived mainly from burning fossil fuels like coal, oil and natural gas. Electrical energy is typically presented as kilowatt-hour (kWh), which is the power multiplied by the number of hours of generation. For example, a 1 GW electric-generation plant operating 100% of a year (365 days x 24 hours/day) would produce 8,760 million kWh of electricity or 8.76 TWh. Wasn’t that fun?!]
Large data centres require enormous amounts of electricity to operate. This electricity is converted to heat in the process of managing the storage and retrieval of data, and in generating content through Artificial Intelligence (AI). For example, “high-end graphics processing units manufactured primarily by Nvidia, dissipate between 300 and 700 watts each”v – and there are millions of these processing units in a large data centre. The Synapse Data Centre is planning a 1.4 GW off-grid natural-gas powered generation facilityvi. This data-centre is ten times the size of any data centre currently operating in Canadavii, and consumes more power than Edmontonviii. More incredibly, the Wonder Valley AI data centre proposed for 26,000 acres near Grande Prairie would require 7.5 GW of power generation capacityix A single inquiry using AI (also known as a large language model) “requires approximately ten times the electricity of a standard Internet search.”x
As electricity is converted to heat energy in data processing, this heat must be removed and dissipated. The most common (and least expensive) method of removing heat from a data centre is through evaporative cooling. Water is pumped alongside the data centre processing equipment to capture the heat being generated. The heated water is then exposed to air using large fans and a portion of the water evaporates. Like a sweating body, the evaporating water removes large amounts of heat energy from the cooling water (that is, evaporation removes the energy added by the data centre processors). As much as 80 percent of the water being circulated evaporates into the atmosphere with the remaining (cooled) water being reused or returned to the environment (contributing to thermal pollution in lakes and rivers, and possibly containing chemicals like corrosion inhibitors).
In northern climates, water will freeze in the winter so it is more common to use a combination of circulating outside air through the buildings to cool the data centre, mechanical air conditioning, and pumping a ‘closed-loop’ glycol-water through the equipmentxi. Keeping the building cool with outside air or air conditioning removes much of the heat generated in data processing, and a glycol-water coolant may be pumped along the data centre processing equipment (as in evaporative cooling), to absorb the waste heat. The coolant is exposed to outside air, moved by large fans, and cooled down. This heat exchange system is not unlike the radiator of your car, in which heat generated by the engine is dissipated by outside cooler air moving through the radiator. Data centres will be built where cooling the data processing equipment is cheapest (i.e. taking advantage of the local climate or inexpensive energy).
The energy required to operate hyperscale data centres is massive, and they generate considerable greenhouse gas emissions contributing to climate change. The Wonder Valley AI data centre, requiring 7.5 GW of natural-gas fired generation would emit 30.5 megatonnes of emissions each year. These additional emissions would reverse the 28 megatonnes avoided through the phase-out of coal-fired generation in Alberta. The first phase of the project, “requiring 1,400 megawatts (MW) of power, would generate about 4.7 megatonnes of carbon dioxide per year using shale gas to run a combined-cycle gas turbine, the most efficient type of gas power. That would make it one of the largest industrial facilities in the province”xii.
The sudden emergence of hyperscale data centres poses a significant challenge to electricity generation infrastructure, energy prices, and competing economic growth sectors. Projections in the United States show data centre electricity demand exceeding 300 TWh by the end of the decade at annual growth rates exceeding 7%xiii. By 2028, data centres are expected to use about 10% of electricity generated in the United States.xiv
Globally, it is estimated that one-tenth of electricity demand growth will be from data centres. “Under the IEA’s central scenario for data-centre growth, the sector’s global electricity consumption would more than double between 2024 and 2030, reaching 945 terawatt-hours (TWh) by the end of the decade”.xv
In 2022, global data centers consumed about 460 terawatt-hours (TWh) of electricity, “accounting for about 1.7 percent of global electricity consumption. It is estimated that data-center electricity consumption will keep growing at an annual rate of 6–22 percent reaching 750–2,300 TWh by 2030. This will lead to annual carbon emissions of approximately 340 million–1,040 million tons, about 0.9–2.8 percent of global carbon emissions based on the 2023 level”xvi. These emissions would exceed those related to air travel.xvii
Energy summary:
– The 38 data centres announced for Alberta are expected to demand over 12 GW of electricity, more than doubling the current electricity consumption.
– The Government of Alberta appears to have a ‘bring your own energy’ policy in which data centres will develop their own off-grid power capacity – mainly using regional natural gas supplies. There does not appear to be any expectation that data centres reduce greenhouse gas emissions through the development of renewable energy generation capacity.
– At 12 GW, using natural gas generated electricity, greenhouse gas emissions will approach 50 million tonnes each year. This would make data centres the second largest emitter (as a sector) in the province, approaching the 87 million tonnes per year from the extraction and processing of bitumen.
– There are opportunities to use this waste heat other industrial process, for district heating (near urban centres), or for heating greenhouses in the winterxviii.
Water Consumption:
It is estimated that the global AI demand for water will be between 4.2 and 6.6 billion cubic meters annually by the end of this decade.
The United States alone used about 65 million cubic meters directly through cooling in 2023, with the expectation that this will double by 2030xix. An additional 800 million cubic meters of water were consumed indirectly through the generation of the electricity required to power these existing data centres. The indirect water usage (for power generation) is approximately twelve times larger than the direct usage (for data centre cooling). At the same proportion, it is estimated that water consumption for data centres will double by the end of the decadexx.
Given these values, it is estimated that the AI model GPT-3 “needs to ‘drink’ (i.e., consume) a 500ml bottle of water for roughly 10-50 responses, depending on when and where it is deployed”xxi. One must hope that the environmental impacts related to consumption of energy and water will be compensated by the benefits of AI-generated cat videos.
Global freshwater scarcity is increasing, with roughly one-quarter of the world’s population facing water stress. “The pattern, on the available evidence, is that the cooling infrastructure for global AI is being built preferentially in regions where freshwater is cheap, regulatory oversight is loose, and the local population is least positioned to negotiate.xxii”
Canada’s National Observer analyzed the locations of 38 data centre campuses proposed in Alberta. Almost three quarters of these proposed data centres are planned for regions under high or extremely high water stressxxiii. [Note: ‘extreme water stress’ is defined by the World Resources Institute as using at least 80 per cent of its available supply. “high water stress” regions use at least 40 per cent.]
Thirteen data centres with a combined power requirement of nearly 6 GW are planned for the Bow and Oldman River basins, which have been closed to new water licenses for two decades.
It should be noted that water footprints of data centres in cold climates may be considerably less that published data which focuses on data centres located in more temperate climates. Closing cooling loops for both primary (data center cooling) and secondary (power generation) processes will likely reduce water consumption compared to cooling processes that rely on evaporation. Media reports suggest the proposed Synapse Data Centre near Olds will employ a closed-loop cooling system.
Material Impacts:
In addition to energy/emissions and water consumption, the third major dimension of AI’s environmental impact is related to its material bast. “The semiconductors, servers, storage systems, and networking equipment that constitute AI infrastructure require a complex array of critical minerals—lithium, cobalt, tantalum, neodymium, dysprosium, indium, gallium, and others—whose extraction involves severe and concentrated ecological damage, disproportionately borne by communities in the Global South.”xxiv
Competition between AI providers requires a rapid cycle of upgrading hardware with more powerful models. These cycles can be from two to three years, generating enormous quantities of electronic waste: “discarded servers, GPUs, memory modules, and networking equipment containing toxic materials including lead, mercury, cadmium, and brominated flame retardants. Global e-waste generation reached 62 million metric tons in 2022 and is projected to grow to 82 million metric tons by 2030. A substantial proportion of this waste is exported, often in violation of the Basel Convention, to processing facilities in West Africa, South Asia, and Southeast Asia, where it is handled under conditions of severe health and environmental risk.”xxv
Endless resource extraction in a finite world, the conversion of living ecosystems into sources and sinks, defines the emerging environmental crisis – climate, hydrology, biodiversity and soil fertility are being challenged by what may be the most resource intensive technology today. As Kate Crawford has written: “AI is neither artificial nor intelligent. Rather, artificial intelligence is both embodied and material, made from natural resources, fuel, human labor, infrastructures, logistics, histories, and classifications. AI systems are not autonomous, rational, or able to discern anything without extensive, computationally intensive training with large datasets or predefined rules and rewards. In fact, artificial intelligence as we know it depends entirely on a much wider set of political and social structures. And due to the capital required to build AI at scale and the ways of seeing that it optimizes AI systems are ultimately designed to serve existing dominant interests. In this sense, artificial intelligence is a registry of power.”xxvi
Policy in Alberta:
According to the National Observer, “Alberta’s Premier Danielle Smith and Prime Minister Mark Carney announced a deal on November 27, 2025, that includes rolling back environmental rules for the province’s data centers. The decision exempts Kevin O’Leary’s massive Wonder Valley data center near Grande Prairie from a provincial environmental assessment. The Alberta environment ministry stated that the project does not require a provincial environmental impact assessment, based on current information about the project.”xxvii
Social Impacts:
Promoters of AI suggest that the benefits to society will out vastly outweigh the costs. This is uncertain – there are many concerns. We have outlined some of these concerns related to the environmental impact of AI data centres: Increasing greenhouse gas emissions at a time when nations have promised to reduce them, towards goals of net-zero by 2050; consuming large quantities of water in water-stressed regions, competing with food production and the health of aquatic ecosystems; and consuming vast quantities of rare-earth metals and other mined resources for products with short life-cycles that will result an electronic waste stream for which there seems little planning to manage.
But what about the social impacts? Why use artificial intelligence?
Artificial Intelligence (AI) represent large language models that quickly access and assess large amounts of data to derive an output to a given query. The general belief is that the more data available, the better will be the AI output. AI data centres have to ‘train’ the language models to make the connections required to provide reliable outputs. It is estimated that ‘training’ is responsible for as much as 80% of the AI energy consumed. To remain competitive, ‘training’ cycles are nearly continuous as more data is ‘scraped’ from any digital source.
To ‘train’ AI models, a dramatic quantity of digital material must be captured – from anywhere, in any way. All data is treated equally, stripped of context, from meaning, and unmoored from intellectual or moral purpose. Kate Crawford asks: “This large-scale capture has become so fundamental to the AI field that it is unquestioned. So how did we get here? What ways of conceiving data have facilitated this stripping of context, meaning, and specificity? How is training data acquired, understood, and used in machine learning? In what ways does training data limit what and how AI interprets the world? What forms of power do these approaches enhance and enable?”xxviii
The data must then be classified by humans for AI training purposes. This requires a team of (possibly unskilled, anonymous) crowd-workers to be tasked in making subjective annotations to data. The combination of indiscriminate selections of data (whether from a peer-reviewed journal or any old ‘slop’ makes no difference) and the reliance of classification through crowd-sourcing and individual designers (who decide the variables and categories in which data is to be classified), raises the risk of bias in the output. In essence, the classifications will reflect the society making the decisions. This bias may range from unintentional discriminatory results to intentional propaganda. Crawford says: “The result is a statistical ouroboros: a self-reinforcing discrimination machine that amplifies social inequalities under the guise of technical neutrality.”xxix Further, the practice of classification becomes a centralizing power: “the power to decide which differences make a difference.”xxx
Kate Crawford concludes: “Artificial intelligence is not an objective, universal, or neutral computational technique that makes determinations without human direction. Its systems are embedded in social, political, cultural, and economic worlds, shaped by humans, institutions, and imperatives that determine what they do and how they do it. They are designed to discriminate, to amplify hierarchies, and to encode narrow classifications. When applied in social contexts such as policing, the court system, health care, and education, they can reproduce, optimize, and amplify existing structural inequalities. This is no accident: AI systems are built to see and intervene in the world in ways that primarily benefit the states, institutions, and corporations that they serve. In this sense, AI systems are expressions of power that emerge from wider economic and political forces, created to increase profits and centralize control for those who wield them. But this is not how the story of artificial intelligence is typically told.”xxxi
In addition to the influence of low-quality data on AI output, and the systemic risk of bias, researchers have voiced concerns about the impacts of AI training that increasingly use the results of its own output in future training – “model autophagia”, where the AI process begins to eat its own body of output. Some are concerned that in the desperation to increase the data available to be used in AI training will result in the increasing use of its own output which, in turn, may lead to a ‘flattening’ of variations of output, eventually resulting in noise. Could AI get dumber in time? Could autophagia result in model collapse?
If AI becomes more effective in improving productivity, replacing white-collar, information-based jobs in the future, what will become of these jobs and the people who rely on them? OpenAI’s Sam Altman predicted (and, after the blowback, un-predicted) a ‘jobs apocalypse’ from the emergence of AI in industry.xxxii The concerns about job-loss in an economy already challenged by stagnation are palpable.
And as economist Paul Krugman has asked, What will AI do to our minds? Using AI to cheat in school is quite possible and likely, but Krugman is more concerned about learning: “students’ use of AI damages their capacity to learn. And what we really mean by learning is the ability to think. Students who rely on large language models to answer questions won’t learn how to think by reasoning through the evidence to form a conclusion. As a result, they will be unequipped to deal with situations in which AI either can’t provide an answer or provides misleading answers. In short, there are good reasons to worry that what we’re calling artificial intelligence will adversely affect the development of our natural intelligence. Moreover, in the case of basic learning, those adverse effects may be virtually irremediable.”xxxiii
Finally, AI-related stocks currently represent about forty percent of the S&P 500 (a market-capitalization-weighted stock index in the United States). There is a lot of pressure for companies to adopt generative AI or be left behind. The incredible growth of AI data-centres will demand a few trillion dollars from tech companies to build infrastructure by 2030 – Where will this capital come from? Will the increased productivity of capital offset the costs of AI? How risk-tolerant will investors be?xxxiv
Summary:
Artificial Intelligence, or large language models, are not generators of neutral data. The approach to ‘training’ these models results in opportunities for bias and control. The reliance of more and more data has raised concerns of ‘model autophagia’ and collapse. Massive financial investments that don’t lead to tangible productivity gains (and profits) may lead to a stock-market collapse. The exponential demand for electricity, water, and mineral resources, not to mention the demand on sinks for greenhouse gas emissions, contaminated water, and electronic waste, may lead to ecological collapse. And the influence of AI systems on learning may lead to a sort of cognitive collapse.
That’s a lot of collapsing. Shouldn’t we have a discussion before we chase this chimera?
References:
i Alberta’s AI Data Centre Strategy, 2024 (https://open.alberta.ca/dataset/f6fe5816-12ac-4ba6-805c-d0a0dd5aebf9/resource/26d62103-ff38-4310-a98f-ab4595a4af74/download/ti-albertas-ai-data-centre-strategy.pdf)
ii White, R. & Bulowski, N. (2026). Three-Quarters of Data Centre Sites Planned for Alberta are in High Water Stress Areas (https://www.nationalobserver.com/2026/03/23/investigations/alberta-data-centres-water-supply)
iii Bakx, K., Young, N., & Ali, R. (2026). Supersized Data Centres are Coming to Canada. One Province is at the Epicentre. (https://www.cbc.ca/news/business/bakx-york-ai-data-centres-alberta-solomon-9.7222388)
iv Electricity Maps (https://app.electricitymaps.com/map/zone/CA-AB/72h/hourly)
v Editors (2026). SpaceDaily (https://spacedaily.com/k-writing-a-single-100-word-email-with-chatgpt-consumes-approximately-the-volume-of-a-standard-bottle-of-water-the-global-infrastructure-processing-ai-queries-is-projected-to-use-the-equivalent-of-hal/)
vi Synapse Data Centre Project Information Package (https://www.olds.ca/media/y3cp0anv/synapse-data-center-project-information-package.pdf)
vii Bakx, K., Young, N., & Ali, R. (2026). Supersized Data Centres are Coming to Canada. One Province is at the Epicentre. (https://www.cbc.ca/news/business/bakx-york-ai-data-centres-alberta-solomon-9.7222388)
viii Anderson, D. (2025). A $10-billion AI data centre races ahead in a rural Alberta town, population 9,679. The Narwhal. (https://thenarwhal.ca/olds-alberta-ai-data-centre/)
ix Lewis, R. (2026). Wonder Valley Advances 7.5 GW AI Campus in Alberta. Calgary.Tech (https://calgary.tech/2026/02/22/wonder-valley-advances-7-5gw-ai-campus-in-alberta/)
x Li, T. (June 2026). The Thermodynamics of Capital: Artificial Intelligence, Energy Crisis and Ecological Crisis. Te Li. Monthly Review. Vol.78, No.02 (pp.17-32).
xi Cadence (2026). Data Centre Cooling Strategies for Cold Climates (https://cadencenow.com/data-center-cooling-strategies-for-cold-climates/)
xii MacPherson, J. (2025). O’Leary’s Gas-Powered Data Centre Emissions Could Wipe Out Albert’s Coal Phaseout Gains. The Energy Mix. (https://www.theenergymix.com/exclusive-olearys-gas-powered-data-centre-megaproject-could-erase-albertas-coal-phaseout-gains/)
xiii 2024 United States Data Center Energy Usage Report. (https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report_1.pdf)
xiv Carbon Brief (2025). AI: Five charts that put data-centre energy use – and emissions – into context. (https://www.carbonbrief.org/ai-five-charts-that-put-data-centre-energy-use-and-emissions-into-context/)
xv Carbon Brief (2025). AI: Five charts that put data-centre energy use – and emissions – into context. (https://www.carbonbrief.org/ai-five-charts-that-put-data-centre-energy-use-and-emissions-into-context/)
xvi Li, T., et. al. (2024). Powering the Data-Centre Boom with Low-Carbon Solutions. RMI (https://rmi.org/resources/powering-the-data-center-boom-with-low-carbon-solutions/)
xvii Giles, M. (2025). By 2030, AI data centres could take a bigger share of carbon emissions than flights do currently. Sherwood News. (https://sherwood.news/world/by-2030-ai-data-centers-could-take-a-bigger-share-of-co-emissions-than/)
xviii Bartlett, C. (2025). Waste Heat Recovery and Utilization Potential from Data Centres in Alberta (https://ucalgary.scholaris.ca/server/api/core/bitstreams/8f66f7)
Xi, Y., et. al. (2023). Waste heat recoveries in data centers: A review. (https://www.sciencedirect.com/science/article/pii/S1364032123006342)
Robinson, A. (2026). Waste Heat from Quebec Data Centre to Grow 80,000 Tonnes of Veggies Per Year (https://www.theenergymix.com/waste-heat-from-quebec-data-centre-to-grow-80000-tonnes-of-veggies-per-year/)
xix Xiao, T., et. al. (2025). Environmental impact and net-zero pathways for sustainable artificial intelligence servers in the USA.(https://www.nature.com/articles/s41893-025-01681-y)
xx Editors (2026). SpaceDaily (https://spacedaily.com/k-writing-a-single-100-word-email-with-chatgpt-consumes-approximately-the-volume-of-a-standard-bottle-of-water-the-global-infrastructure-processing-ai-queries-is-projected-to-use-the-equivalent-of-hal/)
xxi Li, P; Yang, J.; Islam, M.A.; Ren, S. (2023). Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models. (https://www.researchgate.net/publication/369855581_Making_AI_Less_Thirsty_Uncovering_and_Addressing_the_Secret_Water_Footprint_of_AI_Models)
xxii Editors (2026). SpaceDaily (https://spacedaily.com/k-writing-a-single-100-word-email-with-chatgpt-consumes-approximately-the-volume-of-a-standard-bottle-of-water-the-global-infrastructure-processing-ai-queries-is-projected-to-use-the-equivalent-of-hal/)
xxiii White, R. & Bulowski, N. (2026). Three-Quarters of Data Centre Sites Planned for Alberta are in High Water Stress Areas (https://www.nationalobserver.com/2026/03/23/investigations/alberta-data-centres-water-supply)
xxiv Li, T. (June 2026). The Thermodynamics of Capital: Artificial Intelligence, Energy Crisis and Ecological Crisis. Te Li. Monthly Review. Vol.78, No.02 (pp.17-32).
xxv Ibid.
xxvi Crawford, K. (2021). Atlas of AI. Yale University Press: New Haven. (p.9)
xxvii Fawcett-Atkinson, M. (2026). Alberta scraps environmental assessment for Kevin O’Leary’s ‘world’s largest’ data centre. (https://www.nationalobserver.com/2026/04/03/news/alberta-scraps-environmental-assessment-kevin-olearys-worlds-largest-data-centre)
xxviii Crawford, K. (2021). Atlas of AI. Yale University Press: New Haven. (p.84)
xxix Ibid. (p.116)
xxx Ibid. (p.118)
xxxi Ibid. (p.189)
xxxii Murdoch, S. May 26, 2026). OpenAI’s Altman says AI unlikely to lead to ‘jobs apocalypse’. Reuters. https://www.reuters.com/world/asia-pacific/openais-altman-says-ai-unlikely-lead-jobs-apocalypse-2026-05-26/
xxxiii Krugman, P. (June 28, 2026). What will AI do to our minds? Substack post.
xxxiv Krugman, P. (June 13, 2026). Talking witih Azeem Azhar. Substack post.