Canada Now Has Two Data Centre Rulebooks. Only One Can Say No.

I have spent eleven chapters arguing that Canada’s data-centre problem is not a shortage of land, capital, chips or even clean electricity. It is that nobody has assembled those things into a plan. So two plans arriving at once ought to be good news, and in some ways it is. But the interesting question about a policy document is not what it says. It is what kind of thing it is, and what it can compel.

What Ottawa actually published

Canada’s Responsible Data Centre Development Principles set five expectations. Create lasting local benefits. Do not shift electricity costs onto Canadians. Minimise water use and environmental impacts. Be transparent about local impacts. Deliver strategic value to Canada.

Every one of those is the right thing to want, and I would not argue with a word of it. What none of them carries is a number. There is no metric attached to “minimise,” no unit for “transparent,” no reporting cadence, no auditor, and no consequence for a signatory who does none of it.

Compare the instrument, not the sentiment. The European Union did not ask operators to be transparent about water. It required reporting under EU 2024/1364 for facilities above 500 kW and pointed at an ISO standard for how to measure. That is a rule. Principles 3 and 4 are an aspiration, and they land on an industry where, at one Canadian facility I looked at in chapter six, the operator ran for seven years without a water meter. A disclosure expectation presumes the number exists. Meter first, then mandate disclosure.

The signatory list is worth reading closely too. It includes AWS, Google and Microsoft, the three firms that hold roughly 85 percent of Canada’s public cloud market, which is the central problem of chapter seven. It also includes OpenAI and Anthropic, neither of which operates a data centre in Canada at all. What a principle about local benefits and water use asks of a company that runs no facility here is not obvious, and a framework signed by both operators and their tenants needs to say which obligations attach to which.

What Ontario actually proposed

Ontario’s Data Centre Playbook is a different category of object. It would screen projects on three pillars, economic development, digital sovereignty and community benefits, and it proposes moving large new loads into their own electricity rate class, potentially those drawing more than 1 MW, so the cost of serving them is not recovered from everybody else’s bill.

A screening pillar decides who gets a grid connection. A rate class decides what they pay for it. Both are enforceable, because Ontario owns the wire. This is a document that can say no.

It is also not final. The consultation closes on 12 September 2026, and what emerges may be softer, harder or differently shaped. But the direction is clear enough to reason about: Ontario is moving from rationing grid access by scarcity, which is what a queue does, to rationing it by assessed value, which is what a screen does. As I set out in chapter four, that puts it alongside Quebec, British Columbia and Manitoba, all of which now gate large loads in one way or another. What is new is the criteria. No Canadian province has previously screened a data centre on digital sovereignty.

The comfortable reading, and why it is wrong

Put the two documents side by side and they look like one policy in two layers.

Five concerns, two documents, near-total agreement on what matters.

That agreement is real and it is not the point. The two governments are not measuring the same thing.

Ottawa’s objective is denominated in gigawatts. Its national AI strategy estimates Canada needs about 5.5 GW of commercial AI compute by 2030. Its Sovereign AI Compute Strategy put up to $700 million behind commercial data centres, up to $1 billion behind a public supercomputer, and an AI Compute Access Fund that started at $300 million and has since grown. It says it is finalising partnerships proposing 850 MW by 2030, scalable to 2.3 GW.

Ontario’s objective is denominated in cents per kilowatt-hour and jobs per megawatt.

There is no exchange rate between those currencies and no institution empowered to set one. Ottawa can fund, sign memoranda and publish principles. It cannot energise a single megawatt, because electricity is provincial. Ontario controls the wire, carries the ratepayer, and has no stake whatsoever in a national capacity target.

This is why I have stopped calling it a coordination problem. For most of this series I described Canada’s gaps as coordination failures: the ingredients exist, nobody wrote the recipe. That framing implies a missing convener, somebody who could get the parties in a room and find the agreement they both want. Ottawa and Ontario do not need a room. They have different objective functions and non-overlapping powers, and no amount of convening changes either.

An uncomfortable thing about my own numbers

Ontario’s economic-development pillar will be argued with a ratio, and the ratio is one I have spent a chapter defending. Data centres support roughly 0.2 to 0.35 permanent jobs per megawatt. An EV battery plant on the same power supports something closer to eight. Chapter ten makes that comparison at length and I stand behind it.

Applied by a provincial regulator running an economic-development screen, that arithmetic is an argument against approving data centres. My own research, used honestly, works against the federal government’s stated target. I would rather say that plainly than pretend the analysis only cuts one way.

There is a second problem with the pillar, which is that Canada cannot verify it. Statistics Canada has no NAICS code for data centres; the nearest category bundles them with web hosts and one-person consultancies. Every Canadian employment figure in this space is a projection made by the proponent at announcement. A screen applied to proponent projections rewards the proponent who projects most confidently.

What I think happens next

Here is a prediction that can be checked, which is the only kind worth making.

Ontario is proposing to price large loads at full cost causation, possibly from 1 MW upward. Alberta is doing close to the opposite: queue priority for projects that bring their own power, and a levy that falls from two percent for grid-connected projects to one percent for self-generating ones to zero for going fully off-grid. Two provinces, two signals, pointing in opposite directions.

If both regimes hold, the federal 5.5 GW does not fail to get built. It gets built in Alberta, on gas.

Which is the outcome Ottawa’s own second and third principles say it does not want. Cost-shifting is avoided in Ontario by pushing the load somewhere that will absorb it; environmental impact is minimised on paper by a document that does not bind the province where the building happens. The two provincial regimes acting together produce exactly what the federal framework disclaims, and there is no federal instrument standing in the way.

The one mechanism that pays both

There is a version of this that works, and it is not a new idea, just one aimed at the wrong target until now.

Sell faster grid connections in exchange for curtailability. A data centre that agrees to reduce load during system peaks costs the grid far less to serve than one that does not, and the utility avoids building capacity it must then recover from everybody. Duke has demonstrated something like 76 GW of headroom at a quarter of a percent curtailment. Shallow flexibility belongs in the connection agreement; deep flexibility deserves a capacity payment.

Applied to Ontario’s proposal, that means a rate class with a flexible tier rather than a flat penalty on size. It is the only instrument I can find that pays both governments at once. Ottawa gets megawatts. Ontario avoids building peak capacity for a load that could have moved. And the screen improves, because a rate class sorted by megawatts asks how big you are, while one sorted by curtailability asks what you are willing to give back. The second is a much better question.

What to watch

The consultation closes on 12 September 2026, and the shape of what Ontario adopts matters more than the shape of what it proposed. Three things I will be looking for.

Whether the 1 MW threshold survives. It is aggressive: five times lower than the strawman I used when I argued that size is the wrong axis for sorting these projects, and a quarter of Quebec’s 5 MW competitive-selection threshold. A 1 MW inference site in a city that could absorb it tomorrow is not the thing anybody is worried about.

Whether digital sovereignty gets a definition. Ontario’s pillar and Ottawa’s fifth principle both use the word and neither says what it means. I count at least five live definitions now, and chapter seven argues the differences decide the answer. On a residency reading a foreign hyperscaler with a Canadian region qualifies. On a sovereign-cloud reading it does not. A screening pillar that does not pick one will be picked for it, by whoever files the first application.

And whether either government ever attaches a number to a principle. That is the difference between a rulebook and a press release, and after eleven chapters of watching announced capacity outrun built capacity by a factor of sixty, I would rather have one measurable obligation than five admirable ones.

Background and sources for the figures here are in Canada’s Data Centre Race, an eleven-part series on the physical foundations of the AI buildout. The chapters most relevant to this piece are four (power), six (water), seven (sovereignty), ten (alternative uses) and eleven (strategy), all of which were updated on 3 September 2026 to reflect these two plans.