Canada Moves to Tax U.S. AI Models Accessing Public Datasets

For years, Canada has discussed artificial intelligence as though it were a resource that needed to be developed, similar to hydropower in Quebec or oil in Alberta. However, the atmosphere in Ottawa has changed recently. Talent pipelines and compute capacity are no longer the only topics of discussion. It has to do with data. In particular, who has access to Canadian public datasets and should foreign AI firms foot the bill?

Policy circles are beginning to believe that Canada might tax American AI models that use public datasets, especially those from Statistics Canada and other organizations. It is not yet a law. A renewed national AI strategy that sounds suddenly more protective and recent trade tensions have hardened the mood, which is more like a policy drift.

One could almost feel the defensiveness in the air last month as they passed Parliament Hill, with snow packed into the stone steps and staff rushing between buildings. In 2025, Canada repealed its contentious digital services tax in response to pressure from the US. Many people viewed that climbdown, which involved lifting a tax on big American tech companies, as practical. For others, it was a sign of surrender. This new AI tax proposal might be a more subdued, strategic reaction.

On its face, the argument is straightforward. Funded by Canadian taxpayers, Canadian public institutions gather incredibly rich datasets, including health statistics, tax records, and labor market data. These meticulously anonymized and safeguarded datasets are becoming more and more useful for training advanced artificial intelligence models. Shouldn’t Canada receive a portion of the value if American companies are using that data, even if it’s only indirectly through partnerships or licensed access?

Policy analysts at the C.D. Howe Institute have been cautioning that Canada’s AI strategy is devoid of a logical framework for data supply. The nation has made significant investments in compute funds, superclusters, and research chairs. However, structured, reliable access to large-scale datasets is the “missing pillar,” as one recent commentary referred to it. According to this perspective, taxes are more than just a way to raise money. It’s leverage.

However, taxes have symbolic meaning. It alludes to boundaries.

This is a little ironic. Being the birthplace of early deep learning pioneers and internationally renowned research institutes, Canada has long established itself as an open, collaborative AI hub. Evan Solomon has frequently discussed striking a balance between openness and sovereignty in an effort to steer clear of the blatant nationalism that is prevalent elsewhere. There is a noticeable tightening of the language as his public statements change. “Sovereign infrastructure.” “Adoption of AI centered in Canada.” Words that suggest cautious optimism.

Meanwhile, investors don’t seem to know how to interpret this. Some founders from Toronto covertly acknowledge that they rely on collaborations with U.S. model suppliers. It is still too costly to train frontier models from scratch. Startups may find themselves torn between pragmatic necessity and patriotic policy if Ottawa levies fees or regulatory tolls on foreign AI systems that access Canadian datasets.

The privacy issue is another issue that is constantly present. Because of decades of careful stewardship, Statistics Canada continues to enjoy a relatively high level of public trust. That trust may be strained if private companies, particularly those based abroad, are permitted to train models using derived public data. By indicating that such access is subject to monitoring and conditions, a tax could act as an obvious accountability mechanism.

It’s unclear, though, if taxes by themselves address the underlying structural problem. AI models don’t use data in tidy, quantifiable chunks. They consume large, mixed corpora. It could be technically challenging to separate which outputs come from which national datasets. Furthermore, the policy runs the risk of losing its effectiveness and becoming merely symbolic if enforcement becomes unduly complicated.

A larger geopolitical undercurrent is present. When allies impose taxes on American tech giants, the US responds angrily, demonstrating its growing sensitivity to digital taxation. The digital services tax standoff in Canada was a direct example of that friction. Similar conflicts might resurface if access to AI models is taxed, but this time they would be centered on data sovereignty rather than advertising revenue.

However, a change has occurred. The field of artificial intelligence is no longer regarded as a specialty. More than 11,000 people, including academics, artists, and business owners, contributed to Canada’s updated AI strategy during public consultations. That level of involvement indicates that AI is viewed as infrastructure as well as technology. Governance is also invited by infrastructure.

The phrase “How do we catch up?” has given way to “How do we control what we have,” and it’s difficult to miss. It’s a psychological turnabout. It implies that Canada views its datasets as strategic assets, including demographic insights, healthcare usage trends, and longitudinal tax records. In a world where markets are shaped by models that are trained on data, protecting those assets feels more defensive than ideological.

The economic calculus is another. Canada’s prior emphasis on financing computer infrastructure was criticized for being cumbersome and slow. Taxing foreign AI access could be viewed as leveling the playing field, or at the very least, as bringing in money for local reinvestment, if domestic companies find it difficult to create models that are globally competitive. It remains to be seen if that reasoning holds up when examined closely.

It seems like Canada is experimenting in real time as we watch this play out. The nation aspires to be a leader in responsible AI. It seeks the confidence of the public. It seeks financial gains. It also seeks independence without isolation. Those objectives don’t always line up perfectly.

Canada will be signaling something more significant than fiscal policy if it decides to impose taxes on American AI models that access public datasets. It will imply that data is now considered national capital, whereas previously it was considered a neutral public good. Even though that change is slight, it has significance.

Additionally, it will probably proceed more slowly than the technology it aims to regulate, creating questions long before it provides definitive answers, as is the case with most policy experiments in the AI era.