(Version française disponible ici.)

Canada wants to become a global leader in artificial intelligence (AI). But by relying on data centres largely controlled by foreign companies, it risks, above all, funding its own dependence at a cost of billions of dollars.

According to Minister of Artificial Intelligence and Digital Innovation Evan Solomon, the rapid development and adoption of AI are essential if Canada wants to build “the strongest economy in the G7” while “strengthening [its] digital sovereignty.”

To achieve this goal, Mark Carney’s government plans to invest $2 billion over five years to develop a “sovereign” computing capacity. For their part, provinces such as British Columbia and Quebec are seeking to attract the data centres needed to train AI models, drawn by the economic benefits and the demand for electricity.

Such an approach carries a risk: that of leaving AI development in the hands of a few large, mostly American, companies with little tangible benefit to Canada. Prioritizing the energy needs of large data centres also risks undermining energy transition goals by limiting the energy available for other projects.

The strategy of the federal government and the Canadian provinces should prioritize the development of public computing capacity in order to ensure ethical use of AI.

Sovereign computing power

The development of AI, and more specifically of large language models such as ChatGPT, Claude, or DeepSeek, requires significant computing power to analyze and generate data on a massive scale.

According to recent estimates, training a single model requires the use of several thousand state-of-the-art graphics cards over a period of several months.

As computing demand grows rapidly, cloud data centres housing large numbers of graphics cards have emerged as the key infrastructure for training and deploying AI models. At the same time, the offshoring of this infrastructure raises concerns about control over it. In particular, several countries fear becoming dependent on infrastructure outside their jurisdiction and regulatory authority.

In response to these concerns, major players in the AI industry such as OpenAI and Nvidia are promoting the development of sovereign computing capacity. Beyond boosting their graphics card sales and expanding their user base, these companies promise the participating countries greater control over their data and the development of AI. More broadly, sovereign computing capacity is increasingly viewed as a strategic resource in a context of international competition where AI appears poised to play an increasingly significant role.

The Canadian ecosystem

In this race for computing power, Canada is actually doing relatively well, ranking in the top third of countries. Based on our research, Canada is home to at least 285 data centres, primarily concentrated near major urban centres, as shown on the map below.

The largest cloud computing companies, commonly referred to as hyperscalers, together operate nine local cloud regions of Canada in 2025. A cloud region is a specific geographic area within which a cloud service provider operates one or more data centres. This ranks the country 8th out of 47 countries worldwide that host at least one data centre operated by these cloud computing giants.

American domination

Canada, like every other country, remains far behind the United States, which remain the global leader in computing power. According to publicly available data, only nine data centres in Canada have a power capacity of 200 MW, which is considered to be the standard for training the latest AI models.

Furthermore, a single cloud region operated by a hyperscaler features the latest generation of graphics cards produced by Nvidia. For its part, the United States has 18 regions equipped with these chips and accounts for nearly 75 per cent of the computing power dedicated to AI.

Limited economic benefits

However, developing greater sovereign computing capacity does not guarantee economic benefits for Canada. Several studies have pointed out that the construction of data centres has little lasting economic impact on the economies where they are built. They create few long-term jobs and their tax contributions are struggling to make up for rising energy prices caused by the strain they place on the power grids.

The bulk of the value added from these investments remains, moreover, in the hands of the few major U.S. technology companies that dominate the production of graphics cards, the provision of cloud services, and the development of the latest AI models. As such, building computing infrastructure owned by foreign cloud providers does not in itself contribute to the development of AI expertise and knowledge in Canada.

In a context where the profits of these companies are largely shifted to low-tax jurisdictions and where the Canada has abandoned plans to implement its digital services tax – without any alternative multilateral solution yet in place – the net economic contribution of these investments risks remaining marginal.

Environmental cost

At the same time, the environmental impact of AI is growing rapidly. According to the most recent study by the International Energy Agency, energy demand associated with data centres has grown by 12 per cent annually over the past five years and is expected to equal Japan’s annual energy consumption by 2030.

Access to relatively low-cost renewable energy in Canada is particularly attractive to companies seeking to expand their compute capacity and places additional strain on the country’s power grid. In addition to the need to build new, costly infrastructure, this increased demand reduces the amount of energy available for other projects essential to the country’s energy transition – a strain that has notably led Hydro-Québec to double rates for data centres.

In addition to this growing demand for energy, there is also growing pressure on drinking-water supplies, which are used in particular for electricity generation and data centre cooling.

According to a recent study, AI-specialized data centres in the United States are expected to consume as much water as six to 10 million Americans. Although Canada’s colder climate reduces cooling needs, the rapid development of new data centres would intensify pressure on already strained water supplies in a context where droughts are becoming more frequent.

A limited sovereignty

Furthermore, the development of AI-specialized data centres in Canada will have only a marginal impact on achieving true digital sovereignty – the ability to exercise effective control over one’s digital ecosystem.

In practice, data centres operated by major U.S. companies remain under U.S. jurisdiction. Under the Cloud Act, the U.S. government can compel these companies to grant it access to data stored on their servers.

The U.S. sanctions and export control regime gives the United States the ability to restrict or cut off access to essential technology services, including cloud services. Such measures have notably forced the Amsterdam Trade Bank to file for bankruptcy following sanctions imposed in connection with the war in Ukraine.

Control over this computing power would also remain, on a day-to-day basis, in the hands of a few large private companies. As sadly highlighted by the Tumbler Ridge tragedy, there is no evidence that this concentration would ensure the ethical use of AI.

Promoting the development of public AI infrastructure

In this context, the federal and provincial governments should prioritize the development of a public computing infrastructure designed to support non-commercial uses of AI.

Such infrastructure would primarily serve leading research centres such as Mila and the Canadian Digital Research Alliance, as well as higher education institutions and public agencies. The entity responsible for this infrastructure could also establish guidelines governing the development of AI models in accordance with ethical standards and in the public interest. It could also support the research activities of Canadian companies such as Cohere.

Currently, half of the budget for Canada’s Sovereign AI Compute Strategy is intended for the development of public infrastructure. The other half is aimed at supporting private investment. However, investments presented as “sovereign” by large American corporations contribute little to achieving true sovereignty, while fuelling a commercial offering whose social value is often not proportional to the resources it consumes.

The allocation of funding should be reviewed to place greater emphasis on truly public infrastructure and to curb the unchecked expansion of data centres for commercial purposes, along with the environmental consequences that this entails.

In addition to this financial support, provincial governments could contribute by setting aside dedicated energy blocks for such public infrastructure. Finally, the federal government could work with other countries who share similar concerns, with the aim of developing this computing capacity as part of a coordinated effort to minimize its environmental footprint.