 |
| Daniel W. Dylan |
This crash, it is supposed, will result in a correction of sorts triggered by unrealistic expectations, overheated infrastructure spending and the exorbitant operating costs of AI systems. In the United States, commentators have started asking whether national economic growth is now so dependent on AI-adjacent sectors that a downturn could ripple through the entire economy.
Microsoft’s $19-billion Canadian AI infrastructure investment could accelerate innovation, but large AI data centres also come with a heavy environmental footprint. They consume massive amounts of electricity and water, producing significant carbon emissions and local resource strain that Canada will have to grapple with as AI infrastructure expands. Canada likely must ask the same question as commentators have posed in the United States, but with one crucial addition: what would an AI crash mean for Canada’s natural and physical environments?
Tasha Art: ISTOCKPHOTO.COM
AI is not a cloud; it is a resource-intensive industrial system
AI is often imagined as something that happens “in the cloud,” a frictionless digital service; however, as I wrote about in Law360 Canada last year, it is one of the most resource-intensive industries in the modern world. Every chatbot response, every predictive algorithm, every generative image requires the work of thousands of servers running in vast data centres that resemble industrial-scale production sites more than anything digital.
Canada has become a magnet for these facilities because the country’s cooler climate reduces cooling costs. Ostensibly, regulatory stability also attracts investment, and clean hydro and nuclear grids can make AI companies appear more climate friendly. But this growth comes with consequences. That said, some AI data centres may offer some environmental benefits, such as helping utilities to manage electricity demand and to make use of surplus energy that might otherwise go to waste.
AI is consuming provincial energy capacity
A single large data centre can demand hundreds of megawatts of electricity, equivalent to powering a small Canadian city. Provinces that built their grids around predictable industrial or residential loads are now struggling to accommodate sudden surges in demand. In Quebec, for example, narrowing winter capacity margins and rising overall demand are driving increased scrutiny of the grid’s ability to accommodate new large industrial loads. Ontario faces increasing reliance on natural gas generation during a period of higher demand and nuclear refurbishments, raising questions about the province’s emissions trajectory.
British Columbia has taken steps to regulate or limit new large electricity loads from data/AI infrastructure, evaluating how emerging AI and data-centre loads fit within the limits of its primarily hydroelectric system and long-term electrification goals. Meanwhile, Alberta’s deregulated, gas-reliant market positions it as a potentially attractive location for new data-centre investment, even as this dynamic raises considerations around emissions profiles and grid stability.
If the AI boom slows or collapses, these provinces may be saddled with stranded energy assets: expensive substations, transmission lines and generation facilities built to support demand that has evaporated. Sara Blake, for example, recently wrote about a case decided by the Newfoundland and Labrador Court of Appeal that restricted electricity supply.
Data centres also require large footprints, not just buildings, but substations, access roads and transmission corridors. Under s. 35 of the Constitution Act, 1982, and under federal and provincial UNDRIP implementation efforts, governments must ensure meaningful consultation and accommodation with Indigenous Peoples. Yet the speed of AI-focused industrial development that invites Indigenous participation often outpaces the pace of proper engagement. Altogether, carbon emissions also become more concerning as more data centres are built because the rapidly growing energy demand of AI infrastructure increases overall electricity consumption, often outpacing the availability of clean, renewable power.
Industrial competition for water and other environmental issues
Similarly, AI’s cooling needs are enormous. While Canada has abundant freshwater relative to many countries, climate change is tightening water availability in the Prairies and some interior regions. A large evaporative-cooled data centre (which is more energy efficient than mechanical cooling) can consume millions of litres of water per day. If AI growth falters, the worst-case scenario becomes imaginable: water is depleted, unused infrastructure is constructed and ecosystems are disturbed only for facilities to be shut down or abandoned, leaving communities to bear the environmental costs without any lasting economic gains.
Moreover, abandoned or incomplete facilities could litter the Canadian landscape. Decommissioning responsibilities of these facilities might become contested, especially when ownership may involve international firms or complex financing structures. Ultimately, various governments may unwillingly inherit sites requiring remediation under environmental protection statutes. AI hardware is also notoriously short-lived. Chips and servers become obsolete in two to three years, leaving behind toxic materials and metals and other forms of e-waste that are challenging to recycle and must be done in accordance with law. A crash could flood Canadian recycling systems, overwhelming provincial extended producer responsibility (EPR) frameworks.
Utilities that expanded infrastructure anticipating long-term AI demand may face litigation. Ratepayers could be stuck covering the cost of infrastructure built for an industry that has left or disintegrated, and local governments may find themselves responsible for maintaining idle industrial zones, managing stormwater (rain trapped for cooling purposes) from unused sites and addressing contamination without additional tax revenues to do so. Thus, if the AI economy contracts abruptly, environmental liabilities will not disappear. Instead, they will fall to the next responsible party, which is often the public.
A legal framework for AI’s physical footprint is needed in Canada
The Artificial Intelligence and Data Act (AIDA) died when Parliament was prorogued in January 2025. But it was legislation fraught with controversy and problems. AI policy discussions in Canada have focused heavily on privacy, intellectual property and algorithmic governance. These are important, but they overlook the infrastructure that makes AI possible and that which will remain long after the economic cycle has turned. Canada, therefore, needs a coherent, anticipatory legal strategy that includes: a) stronger environmental assessments for data centres and related infrastructure; b) provincial water-use frameworks tailored to high-volume AI cooling demands; c) energy planning that accounts for volatility in industrial demand; d) consultation with Indigenous Peoples; e) mandatory decommissioning and remediation plans before project approval; and, f) national standards for AI-related e-waste and recycling under existing regulatory schemes. Without these safeguards, Canada could be left bearing an environmental mess, while the promised benefits of the AI revolution turn out to be fleeting.
Daniel W. Dylan is an associate professor at the Bora Laskin Faculty of Law, Lakehead University, in Thunder Bay, Ont. He teaches animal law, contract law, evidence law, intellectual property law and Indigenous knowledge governance.
The opinions expressed are those of the author and do not reflect the views of the author’s firm, its clients, Law360 Canada, LexisNexis Canada or any of its or their respective affiliates. This article is for general information purposes and is not intended to be and should not be taken as legal advice.
Interested in writing for us? To learn more about how you can add your voice to Law360 Canada, contact Analysis Editor Peter Carter at peter.carter@LexisNexis.ca or call 905-415-5811.