Canada doesn't have an AI talent problem.
It has a reason-to-stay problem.

It's Canada Day. The barbecues are lit up and down my street, and I'm at my desk instead — because the thing on my mind this July 1st isn't the fireworks, it's why a country this good at making AI founders is this good at losing them. I've spent more than twenty years starting companies from here. I raised roughly $12 million in venture capital, co-founded a company that Zillow acquired and then ran its Platform Services group, founded CTO.ai, and I've spent the years since advising founders doing the same climb. I'm not speculating about the Canadian venture climate from a think-tank chair. I've lived every stage of it — the raise, the exit, the payroll, the board — and I'm writing this on the one day we're all supposed to feel good about building here.

So let me say the uncomfortable part on the patriotic holiday: if a founder asked me today where to incorporate their AI company, the honest default answer is still Delaware — and that's an indictment, not of the founder, but of us. A company can be built entirely in Canada, on research the Canadian public paid for, and the rational move is to make it American before it has a customer. I've watched it happen to companies I backed, and I nearly did it myself. Nobody who does this for a living finds it strange anymore. That numbness is the problem. Because this isn't a talent story or a research story — Canada wins both of those, decisively. It's a story about the boring, unglamorous machinery of scaling a company past its first big round, and that machinery is a policy choice we keep getting wrong.

I think in code, so I wrote the decision out the way I'd write any other config — the inputs that actually move a founder's choice, weighted the way I weight them after twenty years of making this call:

# where do I incorporate kc.io? the honest scoring, from a desk in Canada
def where_to_incorporate(country):
    return sum([
        3 * talent_access[country],       # CA wins: cheap, deep ML talent
        2 * seed_capital[country],         # roughly a tie at pre-seed
        5 * scaleup_capital[country],      # US wins hard: growth rounds live south
        5 * anchor_customers[country],      # US wins: big early buyers are American
        4 * tax_certainty[country],         # CA lost points to 10 months of churn
    ])

# score("CA") < score("US") — and it isn't the talent term that decides it

The talent line is the only one where Canada dominates, and it's outweighed by three terms about money and demand. Read on for why each of those coefficients is what it is.

Canada can build a frontier lab — the proof is already here

Start with the good news, because it's real and it's often buried under national self-deprecation. Canada has produced a genuine frontier-scale AI company. Cohere, based in Toronto, raised $500 million USD at a $6.8 billion valuation in August 2025, then added another $100 million a month later in a second close that pushed its valuation to roughly $7 billion — with AMD, NVIDIA, Salesforce, and Canada's own PSP Investments on the cap table, and Radical Ventures and Inovia leading.

That is not a fluke, and it is not a wrapper. It's an enterprise-grade model lab, headquartered in Canada, competing for the same customers as the labs in San Francisco. It is the existence proof. The lesson I take from it is not "Canada has arrived." It's that the ceiling is real: a Canadian company can raise growth-stage money and build frontier technology from Toronto. The question the rest of this essay is about is why that is the exception and not the pattern — why for every Cohere there are a hundred founders who quietly redomicile before their Series B.

The talent is world-class and it is publicly funded

The top of the Canadian funnel is a genuine national asset, and it exists because the government built it on purpose. The Pan-Canadian Artificial Intelligence Strategy funds three national AI institutes — Amii in Edmonton, Mila in Montreal, and the Vector Institute in Toronto — each eligible for up to $20 million in federal support. These are not marketing shells. Mila is one of the largest deep-learning research concentrations in the world; the modern discipline has roots in labs that these institutes grew out of.

The practical consequence for a founder is that hiring the first ten machine-learning engineers in Toronto, Montreal, or Edmonton is easier and cheaper than doing it in the Bay Area, where a senior researcher costs a California mortgage and turns over every eighteen months. Canadian salaries are lower, the visa situation is saner, and the talent pool is deep and loyal. If your cost structure is dominated by research salaries — and for an AI company, it is — Canada is a legitimately advantaged place to build. The trouble starts when you try to grow.

The middle of the pipeline is where companies bleed out

Here is the part nobody funds a supercluster to fix. Canada has healthy seed capital — angels, accelerators, and a real early-stage venture scene. What it does not have, at anything like the American density, is scale-up capital: the $50-million and $100-million growth rounds that turn a promising Series A into a category leader. When a Canadian company reaches that stage, the money almost always comes from south of the border, and capital has gravity. The lead investor wants a Delaware entity, a US board seat, and eventually a US headquarters "for go-to-market reasons." The company stops being Canadian one term sheet at a time.

The second leak is demand. A startup grows on the back of customers, and Canada is short on large domestic buyers willing to be an early anchor customer for a homegrown AI vendor. Our banks, telcos, and governments are famously risk-averse procurers; they would rather buy a proven American tool than take a bet on a Canadian one. So the Canadian founder's first big logo is usually American too — which means the sales team, then the executive team, then the founder, all drift toward the customers. Talent pulls you to build in Canada; capital and customers pull you to leave. The founders who stay are the ones stubborn enough to fight physics.

And then there's the tax whiplash

If you want to understand why founders hedge by incorporating elsewhere, watch what happened to capital-gains policy. In 2024 the federal government proposed raising the capital-gains inclusion rate from one-half to two-thirds on gains above $250,000 a year — a direct hit to the exact liquidity event a founder spends a decade working toward. Then, on January 31, 2025, the government deferred the change to 2026. Then, on March 21, 2025, Prime Minister Carney cancelled it outright, keeping the 50% rate and raising the Lifetime Capital Gains Exemption to $1.25 million on qualifying small-business shares.

The ending is good. The 50% rate is competitive, and a higher exemption genuinely helps founders. But read the timeline again: announced, deferred, cancelled, all inside about ten months. A founder deciding in early 2025 where to build a company that might exit in 2032 had no idea what their after-tax outcome would be, because the government didn't either. That uncertainty is not neutral — it is itself a cost. Capital and founders price policy volatility the way markets price any other risk: with a discount. When the rules can swing by a third and back in under a year, "just incorporate in Delaware and keep your options open" starts to look less like disloyalty and more like prudence.

Canada doesn't have an AI talent problem. It has a reason-to-stay problem, and reasons to stay are written in statute.

What Ottawa should actually do

None of these leaks are laws of nature. They are gaps in the machinery, and machinery can be built. If the new government means "it's time to build," here is where the highest-impact fixes sit, in order:

1. Stop the policy whiplash and pre-commit. Legislate a stable, competitive founder capital-gains regime and commit to leaving it alone for a decade. The exact rate matters less than the promise that it won't move. Predictability is the cheapest pro-growth policy there is.
2. Underwrite scale-up capital, don't just seed it. Use federal pension capital and a growth-stage co-investment vehicle to anchor the $50M–$150M rounds that currently force founders to raise American. If PSP can back a $7B lab, the model already exists — make it a program.
3. Become the anchor customer. Direct federal and provincial procurement to buy Canadian AI first, with real budgets and short sales cycles. A government that trains the talent should also be its first big logo.
4. Keep the talent tap open. Protect the fast-track immigration lanes that let startups hire globally, and fund the institutes past 2026 so the research pipeline doesn't age out.
5. Tie the incentives to staying. Make the best tax treatment, grants, and procurement contingent on keeping the entity, the IP, and the headquarters in Canada — reward the founders who don't redomicile.
6. Measure the leak, publicly. Track how many venture-backed Canadian AI companies redomicile each year. You cannot fix a pipeline you refuse to instrument.

Canada already did the hard part: it built the talent and proved a frontier company can be born here. What's left is the unglamorous plumbing — growth capital, domestic demand, and a tax code that stops flinching. Get those right and founders stop leaving. Get them wrong and we'll keep doing what we do best: training the world's AI builders, then buying back their companies in US dollars.