Why We Shipped Vinci on the Eve of Canada Day

Nobody takes Canada seriously on AI.

And for most of the last few years, we earned it.

We watched the American labs run the table.

Then we watched China start shipping a serious new model every other week.

And Canada? We produced panels. Funding announcements.

A lot of polite deference to whatever was being built somewhere else.

We were spectators with good manners.

I say 'we' because I was one of them. I'll get to that in another piece.

This one is about what it actually took to stop watching.

And why we picked the day we picked.

The wall nobody talks about

Just a few months ago I went looking for serious GPUs I could actually rent in Canada.

Not in a press release. In production.

On a credit card. That week.

I couldn't find them.

Every path was a dead end, a waitlist, or a sales rep who took the meeting and never called back.

The sovereign-compute story everyone loved to put in a deck didn't exist at the altitude a founder could build on.

You could talk about Canadian AI all day. You just couldn't rent it.

That gap is the part nobody writes about.

It isn't a lack of talent or ambition. It's that the ground floor wasn't built yet.

And it's hard to take a country seriously as an AI power when its own founders have to leave to get a GPU.

So for a while, I understood the shrug. 'Not our fight.'

When you hit that wall enough times, it stops feeling like a stage the country is passing through.

It starts to feel like a fact about the country.

What changed

The compute showed up.

Slowly, then for real.

Canadian sovereign infrastructure I could actually train and serve on.

Hosted here, on Canadian soil, under Canadian rules. Federal support behind it.

Not a slide. Real capacity.

And the moment the compute was real, my excuse was gone.

That's the uncomfortable thing about infrastructure finally arriving.

It takes away the thing you were quietly hiding behind.

I couldn't say 'you can't build this in Canada' anymore.

I could only ask whether I would.

Behind the scenes

What it actually looked like: a tiny team.

Nights and weekends that ran together. A lot of doubt, most of it mine.

We were trying to do something nobody had a recipe for.

Train character and honesty into a model - and have it survive all the way down to something small enough to run on a laptop.

For a long stretch I wasn't sure it would hold at that size.

Small models are supposed to lose the personality first.

That was the bet that kept me up.

No marketing budget. No big launch machine.

Just the question of whether the thing was actually good.

And whether anyone would feel it.

It held.

That's the part I still can't get over.

A 4B that tells you when it doesn't know.

That pushes back instead of flattering you.

That somehow feels like someone instead of something.

Why the eve of Canada Day

We shipped Vinci Piccolo on June 29. Eve of Canada Day. On purpose.

Call it a salute.

I believe some things about this country that the AI conversation has been too cynical to say out loud.

We have the best natural resources on earth.

We have some of the most resilient, level-headed, quietly capable people anywhere.

And we've spent a decade being told the serious technology gets built elsewhere, and we should be grateful to deploy it.

I don't accept that.

I think we deserve an AI champion that actually delivers.

For the people.

On Canadian compute.

Owned in a way nobody offshore can switch off.

Not a research lab that publishes and waits.

Not a deployment shop for someone else's model.

A lab that ships.

Now, let me be honest about the size of this.

Piccolo is a 4B. It won't out-reason the giants, and I'm not going to pretend it does.

It's the smallest version of the idea. The first step, not the climb.

But every lab that ever mattered started with a first model.

This is ours.

And I wanted it to land on the one day of the year we're actually allowed to say we're proud of this place.

The lesson

Here's what I keep coming back to.

A country's reputation in a field doesn't get fixed by panels, funding rounds, or op-eds about falling behind.

It gets changed by someone shipping.

National confidence is downstream of one person - then a few - deciding to build the thing instead of explaining why it can't be built here.

I spent too long treating 'Canada isn't taken seriously on AI' as a fact about Canada.

It's not. It's a fact about what's been shipped so far.

And facts like that have an expiry date. The expiry date is whenever someone decides to do the work.

What I'm watching for next

Whether Canadian compute can carry the heavier models we're building behind Piccolo.

That's the real test, and it's coming over the next few months.

Whether 'made in Canada' can come to mean shipped and relied on - not announced and forgotten.

And whether more Canadian founders decide the wall is a stage to get through, not a verdict to accept.

A year from now, I'd like 'nobody takes Canada seriously on AI' to sound dated.

The only way that happens is if a few of us keep shipping until it does.

Piccolo is live today. Small, free, open, and running on Canadian compute.

Happy Canada Day.

— George