Canada’s AI Roadmap and the Last Gospel of Jobs

Jobs, Jobs, Jobs Is the Wrong AI Strategy

Canada Is Right to Take AI Seriously

Canada’s new AI roadmap is strongest when it stops treating artificial intelligence as a fashionable technology sector and starts treating it as infrastructure. That is the correct frame. AI is no longer just a startup category, a research frontier, or a thing for consultants to staple onto slide decks. It is becoming part of the productive substrate of society: compute, energy, data, talent, public administration, healthcare, logistics, science, and industrial capacity.

That matters because the worst way to think about AI is as a novelty app layer sitting on top of the real economy. That may have been easier to believe when people were mostly using chatbots to summarize emails, generate images, or write politely padded office prose. But that phase is already passing. The more serious version of AI is not a clever assistant in a browser tab. It is a coordination layer, a reasoning layer, an automation layer, and eventually a management layer for increasingly complex systems.

On that front, Canada’s roadmap deserves credit. It recognizes that AI adoption is not optional. If only a small share of Canadian businesses are using AI, then Canada does not merely have a research problem or a startup problem. It has a deployment problem. A country can have brilliant researchers, respected universities, and world-class technical talent while still failing to translate that intelligence into productivity across the economy. The issue is not whether Canada knows what AI is. The issue is whether Canadian firms, hospitals, schools, public agencies, and industrial systems can actually use it.

The sovereignty frame is also correct. Sensitive data, compute infrastructure, public-sector workloads, health systems, research capacity, and strategic decision-making tools cannot be casually outsourced forever. This does not mean Canada needs to build every model, fabricate every chip, or seal itself inside a maple-leaf firewall. But it does mean that dependence has a cost. If the infrastructure of intelligence sits somewhere else, under someone else’s laws, priorities, capital markets, and crisis politics, then Canada’s room to maneuver narrows.

Canada is right to notice that AI sovereignty is not a luxury project. It is what national capacity starts to look like when intelligence becomes infrastructure.

The healthcare emphasis also makes sense. The near-term wins from AI will probably not look like magical robot doctors replacing physicians in one dramatic leap. They will look more mundane, and therefore more useful: reduced paperwork, better triage support, faster imaging assistance, cleaner records, scheduling help, administrative automation, and less friction between patients and care. Canadian healthcare does not need science-fiction spectacle before AI becomes valuable. It needs relief from the slow suffocation of forms, queues, bottlenecks, and overloaded professionals trying to practice medicine while also feeding the paperwork machine.

The same is true of AI literacy. A population that treats AI as either sorcery or apocalypse will not be well prepared to use it. Canadians do not need to become machine-learning engineers. They do need a basic practical understanding of what these systems can do, where they fail, how to verify them, and how to use them without becoming either credulous or terrified. AI literacy is not a decorative education program. It is part of adapting the public to a tool that will increasingly sit inside work, school, government, media, and daily life.

So the roadmap is not misguided at the starting line. It sees several of the right things. AI matters. Adoption matters. Sovereignty matters. Healthcare matters. Public understanding matters. Those are not trivial recognitions.

The problem begins when the roadmap turns from recognizing the machine to measuring its success.

The Last Gospel of Jobs

The weak point is not that Canada wants jobs. Every government wants jobs. Jobs are the safest promise in democratic politics: familiar, measurable, emotionally reassuring, and easy to put on a podium sign. A government can stand in front of cameras and say that a new strategy will create thousands of jobs, and everyone understands the ritual. Work will appear. Wages will flow. Families will be supported. The economy will grow. The future will remain legible.

That language is not foolish. It is inherited.

For most of the modern era, the basic story of technological progress has been simple enough to fit inside a campaign speech: new technology creates new industries, new industries create new firms, new firms create new jobs, and those jobs produce higher GDP. The old jobs disappear, yes, but the economy eventually rearranges itself around new ones. The factory closes. The office park opens. The typing pool vanishes. The software department expands. The pain is real, but the model survives.

AI makes that story less comfortable because it does not only automate muscle. It reaches into administration, coordination, analysis, language, planning, records, code, paperwork, customer support, research assistance, and the countless little cognitive tasks that keep institutions moving. This does not mean all work disappears tomorrow. It does mean the technology points directly at the kind of labor modern economies have spent decades expanding.

That is why the roadmap’s emphasis on job creation feels politically understandable but strategically incomplete. “250,000 jobs” is a comforting number. It tells the public that AI still fits inside the old bargain. The machine arrives, but the wage system remains the main character. Technology changes, but employment remains the proof that society is healthy.

The roadmap has noticed the engine, but it is still reading the old dashboard.

The deeper question is not whether AI will create jobs. It almost certainly will. There will be AI consultants, auditors, infrastructure workers, safety specialists, robotics technicians, data-center staff, trainers, integration teams, compliance experts, and a large supporting ecosystem around deployment. The question is whether job creation is the most important thing AI does.

If AI agents become more capable, the central benefit may not be that they create work. It may be that they reduce the amount of human work required to sustain and improve civilization. That is a much harder sentence for governments to say out loud. It cuts against the moral grammar of the existing economy, where employment is not only a source of income but a source of legitimacy. People work, therefore they deserve wages. Firms hire, therefore they contribute. Governments create jobs, therefore they succeed.

But what happens if productive capacity begins to detach from human headcount?

This is where the employment frame starts to strain. Employment has historically been more than a labor arrangement. It has been the main distribution mechanism. Most people receive purchasing power because the economy needs their time, attention, skill, and endurance. If AI and robotics weaken that link over time, then the policy question changes. The issue is no longer simply how many jobs can be created around the new technology. The issue is how society distributes the benefits of production when less human labor is needed to produce them.

That is not an argument for ignoring employment. People still need income, purpose, structure, and social participation. A sudden labor shock would be destructive if governments treated it as an abstraction. But clinging to job counts as the master measure of AI success risks missing the larger transition. It keeps the conversation trapped inside the old formula at the exact moment when the formula may be losing explanatory power.

If AI works, the question is not how many jobs can be wrapped around it. The question is how much civilization can run with less forced human labor inside the machine.

A more serious AI strategy would still care about workers, but it would measure something deeper than headcount. It would ask how much administrative burden can be removed from healthcare. How much permitting can be accelerated. How much logistics can be optimized. How much public-sector backlog can be reduced. How much scientific work can be amplified. How much the cost of necessities can be driven down. How much resilience can be built before demographic aging, fiscal pressure, and labor shortages squeeze the system harder.

Jobs matter. But if AI becomes a general-purpose layer of automation and coordination, then capacity matters more.

That is the real critique of Canada’s roadmap. It is not that the government failed to chant the correct innovation slogans. It is that the roadmap still treats employment as the highest proof of success, when the more important test may be whether Canada can build enough AI-enabled capacity to remain prosperous, resilient, and self-directed in a world where labor is no longer the central constraint.

Sovereign Compute Means Deals, Not Slogans

Sovereignty is easy to invoke and hard to build. It sounds clean in a strategy document because the word itself does so much emotional work. It suggests control, maturity, independence, seriousness, and national purpose. But in AI, sovereignty cannot mean isolation. Canada does not need to build every model, own every chip fab, operate every data center, and recreate the entire frontier stack behind a maple-leaf firewall.

That would not be sovereignty. It would be expensive theater.

The real question is more practical: what does Canada need to control, guarantee, reserve, or negotiate so that it can still act when compute becomes scarce, expensive, politicized, or prioritized elsewhere? A country does not need to own the whole ocean to avoid drowning. But it does need ships, ports, maps, fuel, and a claim on passage when the weather turns.

This is where Norway offers a useful comparison, with one important caveat. Stargate Norway was originally announced by OpenAI, Nscale, and Aker as a major AI data center project near Narvik, with 230MW of initial capacity, ambitions to expand by another 290MW, and a target of 100,000 NVIDIA GPUs by the end of 2026. The announcement emphasized the familiar northern advantages: abundant hydropower, low-cost energy, cool climate, and existing industrial capacity. It also included priority access language for Norway’s AI startups and scientific researchers, with surplus capacity serving wider regional demand.

But the project also shows why Canada should focus on structure, not logos. OpenAI’s role did not remain fixed. The Narvik campus later became tied to a major Microsoft-Nscale agreement, with Microsoft securing additional high-performance AI compute capacity at the same 230MW site. That shift matters, but it does not weaken the lesson. It clarifies it.

The lesson is not “Canada should copy Norway and call OpenAI.” The lesson is that sovereign compute is built through practical arrangements: major AI firm or hyperscaler, domestic industrial partner, abundant clean power, cold climate, large-scale infrastructure, priority national access, and regional export capacity. The brand name can change. The structure is what matters.

Canada has many of the same ingredients. It has hydropower, cold climate, land, political stability, technical talent, proximity to the United States, provincial energy systems, pension capital, public research institutions, and a real need to prevent its AI future from becoming a rented service controlled elsewhere. British Columbia, Quebec, Manitoba, Newfoundland and Labrador, and parts of the North all contain pieces of the compute geography puzzle. The question is whether Canada can assemble those pieces before the next decade’s infrastructure map hardens around other countries.

This is where the roadmap should become more concrete. A public supercomputer may be useful. Research funding may be useful. AI literacy may be useful. But none of that substitutes for large-scale compute capacity under reliable Canadian access conditions. If Canada wants sovereign AI, it should be asking a blunt infrastructure question: how many major AI data centers can the country attract, under what terms, with what guaranteed domestic access?

The terms matter. Canada should be willing to invite the big players in, but not as a passive landlord grateful that someone noticed the weather. If a global AI firm or hyperscaler wants Canadian land, power, permitting, cooling, grid access, public stability, and proximity to North American markets, then Canada should negotiate like those things have value. A share of the resulting compute should be reserved for Canadian public institutions, universities, researchers, startups, healthcare systems, and emergency national use.

That bargain does not have to be hostile. It does not require chest-thumping nationalism or suspicion of every foreign partner. It simply recognizes that infrastructure built inside a country should strengthen that country’s capacity to act. Data residency rules, crisis access provisions, heat-reuse requirements, energy transparency, provincial participation, and Canadian co-ownership where practical should be part of the conversation from the beginning, not polite afterthoughts once the ribbon-cutting photos are taken.

Sovereignty is not a press release. Sovereignty is capacity you can still use when everyone else is trying to save themselves.

That is why the Norway example is useful. It translates “sovereign AI” from slogan into structure, while also showing that these arrangements can shift quickly among the major players. The real issue is not whether the data center carries the right launch-day branding. The issue is whether the arrangement leaves the host country with meaningful access, durable infrastructure, domestic capability, and strategic room to maneuver. If compute becomes one of the central inputs of national production, then Canada cannot afford to treat it like ordinary cloud procurement.

A country that waits until a crisis to secure compute will discover what every country discovers too late about strategic dependencies: the market is friendliest when you do not desperately need it. During a smooth decade, foreign compute looks efficient. During a hard decade, it may become expensive, rationed, politically conditional, or simply unavailable at the scale required.

Canada does not need to build a Canadian OpenAI from scratch to take AI sovereignty seriously. It needs to make sure that when intelligence becomes infrastructure, the infrastructure is physically present, politically negotiable, and practically available inside Canada.

Otherwise, “sovereign AI” risks becoming one more slogan printed on top of someone else’s machine.

The Real AI Question: How Many Agents Can Canada Deploy?

If AI remains a modest productivity tool, then the roadmap’s job language may be enough. In that version of the future, AI becomes another layer of software. It helps workers move faster, firms become more efficient, hospitals reduce some paperwork, and governments modernize a few tired systems. Jobs are created around integration, training, compliance, auditing, consulting, and infrastructure. The old economic story bends a little, but it does not break.

But if AI agents become more capable, the question changes.

At that point, the issue is no longer simply how many people can be employed in the AI sector. The issue is how much of the country’s operating burden can be shifted onto systems that do not sleep, retire, burn out, commute, forget forms, or spend three weeks trying to find the right department. Canada should not only be asking how many jobs AI will create. It should be asking how many useful agents can be deployed across the systems that already struggle to function.

How many agents can help run public administration? How much paperwork can be removed from healthcare? How much permitting can be accelerated without turning oversight into a rubber stamp? How much logistics planning can be automated? How much municipal, provincial, and federal administrative drag can be reduced? How much routine coordination can be taken off human beings who are currently drowning in inboxes, forms, compliance loops, scheduling conflicts, and ancient software that looks like it was last blessed by a fax machine?

This is where AI becomes more than a productivity story. It becomes a state-capacity story.

A serious AI strategy should be asking how much friction can be removed from the systems that keep daily life expensive, slow, and brittle. If housing takes too long to approve, if medical records do not move cleanly, if public benefits are hard to access, if infrastructure projects crawl through paperwork, if small businesses lose hours to administrative nonsense, then the problem is not only labor supply. It is coordination failure. It is institutional drag. It is civilization spending too much of its energy proving to itself that a form has passed through the correct sequence of tired hands.

AI will not fix that automatically. Bad systems can automate their own dysfunction. A broken process with an AI layer on top can simply become a faster broken process. But that is precisely why the strategic question matters. Canada should not treat AI deployment as a vague business adoption target. It should identify where intelligent automation can reduce real bottlenecks, lower real costs, and make public systems feel less like hostile architecture with a login page.

Jobs are a comfort metric. Capacity is the survival metric.

This becomes more important if labor-market disruption accelerates. A government that only prepares for AI as a job creator may be caught flat-footed if AI becomes a labor reducer. The issue would not be that work vanishes overnight. The issue would be that the old connection between employment, income, production, and social legitimacy starts weakening faster than institutions can adapt.

In that world, the central policy question becomes uncomfortable but unavoidable: can AI drive enough productivity and cost deflation to make a more generous distribution system viable if employment becomes less reliable as the main route to income?

That is the bridge Canada’s roadmap has not fully crossed. It talks about adoption, jobs, literacy, investment, and sovereignty. Those are important. But beneath them sits a larger possibility. If AI and robotics can reduce the cost of administration, healthcare support, manufacturing, logistics, education, food production, and basic services, then the country’s long-term goal should not merely be to attach more jobs to the machine. It should be to make life cheaper to sustain.

That does not mean treating people as obsolete. It means refusing to confuse human dignity with compulsory participation in every inefficient process the old economy happened to require. There is nothing sacred about waiting on hold, manually re-entering the same information into five systems, burning medical time on paperwork, or delaying housing because approval chains move at the speed of institutional molasses. If AI can remove that burden, the victory is not that someone got a new job babysitting the spreadsheet. The victory is that the burden shrank.

A country preparing for serious AI should not only count workers. It should count agents, compute, energy, automated workflows, and the falling cost of staying alive.

This is where sovereign compute returns to the center of the argument. Agents do not run on vibes. Automation does not scale on press releases. If Canada wants AI to improve healthcare, reduce administrative drag, support scientific research, strengthen public services, and prepare for deeper economic disruption, it needs the physical capacity to run those systems. Compute is not separate from the post-labor question. It is one of the conditions for answering it.

If the labor market is only modestly disrupted, Canada’s roadmap may be adequate. But if AI agents and robotics keep improving, the country will need a more serious scoreboard. How much productive capacity can be automated? How much of that capacity is under Canadian control? How much can the cost of essentials fall before social strain outruns policy imagination? How quickly can the state respond if the old labor bargain weakens?

Those are not science-fiction questions anymore. They are the questions hiding underneath every government document that still promises jobs while quietly preparing for automation.

Canada does not need to panic. But it does need to look directly at the machine it is inviting into the economy. If AI becomes a general-purpose layer for cognition, coordination, and eventually physical automation, then the countries that do best will not be the ones that produce the most comforting employment slogans. They will be the ones that understand what the technology is actually for: increasing useful capacity, lowering necessary costs, and giving society more room to maneuver when the old assumptions start to fail.

Canada Needs a Better Scoreboard

None of this means Canada’s AI roadmap is a failure. It is not. A bad strategy would ignore AI, reduce it to a startup subsidy program, or treat it as a passing software trend that can be managed with a few grants and some reassuring language about innovation. Canada’s roadmap is more serious than that. It understands that AI touches sovereignty, productivity, healthcare, research, public capacity, and national resilience.

That is a good starting point.

Canada also has real advantages. It has energy. It has cold climate. It has research talent, stable institutions, immigration appeal, public healthcare systems under visible strain, proximity to major markets, and a political reason to care about sovereignty more intensely than it did a decade ago. It has provinces with serious electricity assets. It has pension capital that understands long-term infrastructure. It has universities and public research institutions that can absorb compute if the country makes enough of it available. These are not small things. They are the beginnings of a national AI posture, if Canada chooses to assemble them into one.

But the roadmap still needs a better scoreboard.

It is not wrong to care about jobs. People need income. They need security. They need dignity, purpose, routine, and some confidence that the future being built will still include them. Any government that talks about AI without talking about workers will sound bloodless, and probably deserves to. The problem is not that Canada mentions jobs. The problem is making job creation the central proof that the AI strategy is working.

That metric belongs to an older economic world.

The next version of Canada’s AI strategy should be less attached to the old gospel of job creation and more focused on strategic capacity. How much compute does Canada control or guarantee? How much energy can be directed into productive automation? How many agents can be safely deployed across public and private systems? How much administrative drag can be removed from healthcare, permitting, logistics, education, and government services? How much cheaper can the necessities of life become if automation is aimed at real bottlenecks instead of decorative novelty?

These are not softer questions than jobs. They are harder ones.

A country can create jobs while remaining dependent, slow, expensive, and brittle. It can also reduce unnecessary labor while becoming more capable, more resilient, and more humane. The difference is whether policy is measuring the number of people kept busy or the amount of useful work society can accomplish with less strain.

That distinction will matter more as AI improves. If intelligence, automation, and production become increasingly decoupled from human labor, then national resilience will not come from pretending the old bargain is untouched. It will come from building the infrastructure, institutions, and distribution systems needed for a different kind of economy. One where work still exists, but no longer carries the entire burden of income, legitimacy, production, and social order.

Canada’s AI roadmap is a first draft written partly in the language of the old economy. That is understandable. Governments know how to promise jobs. They know how to announce funds. They know how to speak of growth, training, and competitiveness. But AI may force a harder vocabulary on them. The countries that navigate the next decade best will not be the ones that preserve the most familiar slogans. They will be the ones that build enough compute, energy, automation, and institutional imagination to remain sovereign when labor, production, and distribution begin to uncouple.

That is the opportunity still sitting inside Canada’s roadmap. It has many of the right pieces. It has the outline of seriousness. It has noticed that AI is infrastructure. Now it needs to follow that thought all the way down.

Canada does not need an AI policy that chants “Jobs! Jobs! Jobs!” at the edge of the post-labor machine. It needs a compute-and-capacity strategy serious enough for the world that machine may create.

- Iarmhar

June 8, 2026