In most parts of the world, three firms - Amazon, Microsoft, and Alphabet - dominate the public cloud market.

In Canada, they control 85% of the market. The number is similar across Europe. Globally, it’s around 70%. Throughout the democratic world, the infrastructure that governments, hospitals, banks, and startups all run on is owned by a handful of companies headquartered in one place.

When a few providers hold the infrastructure your economy runs on, you tend to discover the dependence at the exact moment you'd most want a way out — a price change you can't refuse, terms you can't push back on, data you can't practically move. 

Substitute your own country's name, and the story barely changes. And when even Signal, a company built to be independent, concedes there is no alternative to the hyperscalers, you're not looking at a preference. You're looking at structure.

This isn’t just a problem for countries. Consider an American city facing the renewal of its cloud services agreement;  it can’t meaningfully negotiate and can’t realistically escape. Or a Latin American regional health service that would need a year and millions of dollars it doesn’t have to migrate its data to a different cloud provider. Or a Fortune 500 company's finance team that long ago accepted "egress fees" as the cost of doing business. 

They may call it vendor lock-in or budget capture, but the central problem is the same.

Exit is what disciplines power.

What everyone of them actually wants is autonomy. Not to be at the mercy of someone who can squeeze you — even if they never get around to squeezing. A landlord who can raise your rent to anything they choose is still your landlord, whether or not they do so. And the thing that turns a market defined by dependence into one defined by choice is liquidity — how cheaply and quickly you can switch providers. Exit is what disciplines power.

This is the argument Curtis McCord and I make in Parting Clouds: Creating a Competitive Marketplace for Compute, a new report from the Canadian Anti-Monopoly Project. 

Ownership asks: Can we build our own? Autonomy asks: Can we leave if we need to? 

We wrote it about Canada because that's our home turf  — but almost nothing in the diagnosis is uniquely Canadian, and that's the point. A sovereignty strategy can ask one of two questions. Ownership asks: Can we build our own? Autonomy asks: Can we leave if we need to? 

Most of the energy in this debate goes to the first question. We think the second is the more honest and more winnable one for a mid-size country, a region, or a firm. Better still, getting it right makes the first option easier to execute.

We believe a combination of competition policy and public procurement can help reshape the market for compute. The goal isn't to evict the hyperscalers; they aren't going anywhere, and for plenty of workloads, they're genuinely good at what they do. 

The goal is to make the hyperscalers interchangeable and, where it makes sense, replaceable with a wider range of alternatives, including local providers, public infrastructure, cooperatives, and organizations’ own systems. Standards are what make that switch feasible rather than theoretical.  

Think of the electrical grid: municipal utilities, investor-owned utilities, and farmer-owned rural cooperatives all feed the same wires because they meet the same standards — voltage, frequency, interconnection. The point isn't that ownership stops mattering; it's that the standard guarantees make it easier to connect and disconnect not just to different providers but also to different production models. That's what we want for compute.  

So how do you make compute portable? The report lays out three levers that have to work together. 

Canada's federal government alone spent over $156 million in a single year, and it treats buying power as an accounting line item instead of what it is: the power to set the market's defaults. 

The first is procurement. Governments are among the largest cloud buyers on the planet — Canada's federal government alone spent over $156 million in a single year, and that's a small economy — and they treat that buying power as an accounting line instead of what it is: the power to set the market's defaults. 

Governments should buy cloud only from providers, domestic or hyperscaler, that meet existing portability standards. Don't wait to legislate elegant new ones; use what the market has already converged on, and make portability the price of the contract. 

And here's the part worth sitting up for: a city or a region doesn't have to wait for its national government to act. Sub-national and regional governments are already where digital policy gets made when national legislatures stall — privacy, right-to-repair. A handful of big public buyers writing the same clause would move this market faster than any statute.

The other two levers are familiar. Rules that treat cloud as a utility — banning egress fees, prohibiting self-preferencing, and forcing billing transparent enough to compare two providers. And competition enforcement aimed at the machinery of lock-in: bundling, tying, predatory credits, and discriminatory licensing. Both necessary. Both slow. It's procurement that serves as the carrot and stick.

Such a strategy gets stronger with collective action. No single mid-size country can force a market this concentrated to standardize. Neither can one city, nor one company. A mid-size democracy has the sovereignty motive, but not enough buying power.

Collectively, the interests of mid-size democracies, large regions and public systems, and big companies are not all the same, but they are aligned and form a coalition that hasn’t noticed itself yet.

Alone, a large region or public system has insufficient buying power and treats this as someone else's fight. A big company has both the motive to avoid dependence and the lack of coordinating power, as well as a public language to advocate for it. Collectively, their interests are not all the same, but they are aligned and form a coalition that hasn't noticed itself yet.

This gets more urgent, not less. The same firms that own the cloud are the ones financing and hosting the AI labs building on top of it. Commoditize the base of the stack now, while it's still about storage and bandwidth, as this might make it easier to find alternative models for governing and managing intelligence itself.

Not every country, jurisdiction, or company can own its own cloud. But you should have the ability, at a minimum, to leave a supplier and, ideally, the capacity to create an ecosystem that fosters autonomy and agency. There are no easy paths to such a future, but we think this one is the most realistic. 

Parting Clouds can be downloaded here. If you're working on this somewhere — a ministry, a city hall, a procurement office, a company — we'd love to compare notes.