Artificial intelligence may seem like an entirely digital technology, but its function depends on very real infrastructure with a physical footprint. Across Canada, proposals for industrial-scale data centres are multiplying as stakeholders promote AI development and companies race to build the computing power needed to train more sophisticated models. 

Supporters point to innovation and economic potential from Canada becoming a global AI leader. Critics raise questions about electricity demand, water consumption, land use, environmental impacts and who ultimately benefits from AI and related infrastructure projects. 

Dr. Anne Pasek, Canada Research Chair in Media, Culture and the Environment at Trent University, is one of the few researchers in Canada studying the politics and governance of data centre development and helping bridge the knowledge gap among stakeholder groups in the AI arms race. In this Q&A, Professor Pasek explains the role of data centres in AI, how projects are approved, and what questions citizens need to be asking about data centre development projects. She also shares how her research has supported the development of Canada’s first-ever public toolkit alongside the Council of Canadians and collaborators across the country to help communities play an active role in determining where and how the cloud meets the ground. 

Why are data centres such a hot topic in conversations about AI?

People often talk about "the cloud" as though our emails, movies and AI chats exist somewhere up in the sky. Everything we do online depends on physical infrastructure on land. Anything that stores or processes data for people to access elsewhere is technically a data centre. An old computer in someone's basement hosting a website counts.  

The conversation right now is about the rapid expansion of industrial-scale facilities built specifically to support artificial intelligence. They're giant warehouses filled with computer servers — like a Costco with server racks – to support the enormous amounts of computing power required by AI. The facilities can consume as much electricity as a small city, with some using more than a million gallons of water every day for cooling.  

Canada currently has five hyperscale data centres—facilities requiring at least 50 megawatts of power—but nearly 100 more projects are proposed or in development.  

The physical infrastructure behind our digital lives has always existed, but AI is driving demand to a completely different industrial scale. These facilities are exponentially larger, more energy-intensive, and are being rushed through planning and consultation processes. 

Who's building these facilities, and why are companies investing so heavily? 

Many people assume companies like OpenAI or Microsoft are building these facilities themselves. Sometimes they are, but increasingly they aren't. Many proposed Canadian data centres are being developed by infrastructure or asset management companies that act more like landlords. They build facilities hoping technology companies developing large AI models will eventually lease computing space. 

OpenAI, for example, rents computing capacity from partners rather than operating its own network of data centres. Their entire approach is based on scaling—throwing as much computing power as possible at increasingly sophisticated models. There isn't a single facility in the world that's large enough to support all of that work, so companies are constantly looking for additional computing capacity. 

That's one reason we're seeing what many describe as a gold rush for AI infrastructure. It’s a volatile economic bet. 

Why is there this exponential interest in developments in Canada? Does it matter where AI data centres are built? 

Canada offers relatively abundant land, electricity and water compared to many jurisdictions where these companies have already built extensively. 

Historically, data centres needed to be located close to people because they stored movies, websites and other content where speed mattered. You don’t want Netflix buffering halfway through a show, so for streaming stuff, there's multiple copies of all these media files living in different data centres in different parts of the world to keep things running fast in the region. 

That doesn't matter as much for AI model training, since it's something that can take months to achieve. You can build those facilities in the middle of nowhere, get them to execute their tasks, and then dribble the results down a much slower internet pathway when it's done. You'll still need data centres that are close to people for doing the 'inference' part of AI (i.e. the back and forth between a user and a trained model). Peterborough doesn't need a data centre with AI models in it in the city to run ChatGPT, but it does need one somewhere in the general region. OpenAI can train the next ChatGPT model (or just retrain an old one) pretty much anywhere it likes. 

What case is being made for these developments? 

Data centre proposals are typically framed around economic development, jobs, innovation and Canada's competitiveness in AI. Those benefits may exist to an extent, but this is an area where more scrutiny is needed. 

Most employment is temporary, and created during construction. Once operational, even very large facilities typically only employ between 20 and 100 permanent workers, such as security personnel, while much of the specialized technical work and equipment procurement happens elsewhere. 

Communities therefore need to ask whether the promised economic benefits justify the long-term demands on electricity systems, water supplies and local infrastructure and landscapes. 

What questions do you think are missing from the public conversation? 

Right now, governments are moving quickly to encourage AI development, but policy and public understanding haven't kept pace. Much of the discussion focuses on attracting investment rather than asking what responsible AI infrastructure looks like. That means, for conversations about data centres which are needed to support the AI arms race, we need to ask questions about electricity demand, water use, greenhouse gas emissions, land-use planning, biodiversity and who ultimately benefits from these projects. 

When these projects connect to our grids, and spur new upgrades to our electrical system, these costs can be passed down to everyday consumers, raising our electrical bills. 

To avoid this, and to scale faster, many proposals now also include on-site methane gas generation.  This is a trade-off that’s quite bad for local air pollution and global climate change. 

Responsible AI isn't only about ethical algorithms. It's also about the ethics of the physical systems that make those algorithms possible. 

How are these projects reviewed and approved? 

There isn't a single approval process. Municipal governments typically oversee zoning, planning approvals and local infrastructure. Provincial governments play important roles through electricity planning, environmental regulation and sometimes water permitting. Federal governments shape broader AI and industrial policy. 

Because responsibility is spread across different levels of government, it can be difficult for communities to understand where decisions are actually being made. 

Hamilton has recently become an important example. The city approved a one-year moratorium on new data centre development to give policymakers time to develop clearer rules before additional projects move forward.  

The toolkit we’ve developed highlights moratoria as one approach communities can use to give governments time to catch up with rapidly changing technology.  

What role can communities play? 

Residents can attend planning meetings, review environmental assessments, speak with elected officials and ask questions about electricity demand, water consumption, tax agreements, environmental impacts and long-term community benefits.  However, people need access to reliable information and meaningful opportunities to ask questions before decisions are made. That isn’t always guaranteed when projects are rushed through approvals processes, or when non-disclosure agreements prevent important information from coming to light. 

You've helped develop Canada's first public toolkits on data centre development. What prompted that work? 

Across Canada, communities are suddenly finding themselves trying to understand proposals involving hundreds of megawatts of electricity, millions of gallons of water and complex planning processes. 

Most people don't have backgrounds in energy systems, planning or AI infrastructure, yet they're expected to participate in decisions that could reshape their communities. The toolkit aims to help bridge that knowledge gap. It explains how data centres work, how projects move through municipal and provincial approvals, what questions communities should ask, and how people can get involved in these discussions and decisions. 

The toolkit isn’t about telling communities whether they should support or oppose a project. It's to help ensure everyday people have a fair say in decisions about data centre development and understand the trade-offs.  

Looking ahead, what questions should Canadians be asking about AI? 

We're having lots of conversations about digital AI tools, but far fewer about the physical infrastructure that makes those tools possible. 

Canadians should be asking what kind of energy system we want to build. How should communities weigh economic development against environmental costs? What kinds of regulations are needed before AI infrastructure expands even further? 

The future of AI is software, as well as energy, water, land, governance and public accountability. Understanding those connections is essential if we want AI to develop in ways that serve people, communities and entire countries. There’s more than one way forward here, and everyday people can help shape our national technological trajectory by engaging in their local infrastructural debates. 

Learn more about the School of the Environment and Cultural Studies at Trent University.