Four stories about Canada’s grid turned out to be one argument, and the thing we keep failing to build isn’t a wire or a battery. It’s the market that would pay for them.

Over the past month I have written four articles about what I took to be four different problems. A transmission gap between provinces. A pumped-storage puzzle Canada solved in 1957 and then forgot. A hundred-year-old power plant whose cost overrun nobody remembers. And an AI data-centre boom threatening to draw more power than the province I live in has ever consumed.

Somewhere in the fourth piece, I realised I had been writing the same article four times.

Each story arrived dressed as a developer’s question. How to stitch thirteen grids into one. How to store electricity for the calm week in January. How to build century-scale machines using honest accounting. How to power a data centre that can never go dark. And in each one, the building turned out to be the settled part. What was missing was never the technology. It was the set of rules that decides what gets valued, who pays, and who carries the risk of being wrong.

We argue endlessly about what infrastructure costs. We almost never argue about what it buys. And we have built, in this country, an almost perfect machine for confusing the two.

The trap in the century argument

Consider how we talk about a big project. The terrace above the Niagara Gorge holds two answers, side by side.

The hundred-year-old generating station there was estimated around twenty million dollars and finished at eighty-four, in the dollars of the day; the overrun triggered a provincial inquiry and nearly destroyed the man the plant is now named after. It passed its hundredth birthday still running, and nobody remembers the scandal.

Its neighbour, the pumped-storage station commissioned in 1957 to shift coal-fired baseload into the evening peak, I could not price at all. Its construction cost appears in no record I searched, and that absence is the point: nobody kept the outrage alive. But the machine was still quietly pumping water uphill last night, on a grid its designers would not recognise, doing a job they had never heard of: balancing wind and solar.

Cost divided by an election cycle looks like a scandal. Cost divided by a century looks like a bargain.

That is a real insight, and it is also a trap, because it can be used to wave away every overrun as tomorrow’s inheritance. So let me be precise about where the century argument holds and where it falters. An asset that outlives its original purpose is a gift only if someone was charged an honest price for it in the first place. The century denominator justifies the value. It does not excuse a dishonest cost.

Which is why the more useful question is not what these things cost. It is why our system is so bad at pricing what they are worth.

One root, three blind spots

Here is the pattern underneath all four stories. Our electricity markets are very good at pricing a kilowatt-hour delivered right now, and much worse at pricing anything else. They pay for capacity, but not for duration; a four-hour battery and a week of stored water clear the same procurement. They can price electricity flowing across a provincial border, but not the cost of building the line that carries it. And they are only now deciding what to charge a class of customers asking to add more load than the system has ever carried. Three blind spots, one root.

Duration.

Canada has roughly 1,200 mapped pumped-hydro sites and enough theoretical potential to make the number almost meaningless. The physics is a century old. Yet we have not built a pumped-storage facility since 1957, and the reason is not geology. It is that an energy-only market like Alberta’s pays handsomely for the electron you deliver at six o’clock and almost nothing for the ability to deliver it on the fifth frozen, windless evening of a January cold snap. The market rewards the daily cycle and ignores the thirty-seventh hour.

Patient capital, much of it Canadian pension money, already buys eighty-year storage assets, but it buys them abroad, where someone has written a contract that pays for duration.

The capital is not waiting for better technology. It is waiting for a market that prices what the technology does.

Cross-border cost.

Now widen the lens to transmission. Thirteen governments would have to cooperate to build an east-west grid. The federal power to designate an interprovincial line has existed for decades and has never once been used. When Ottawa finally moved this year, the benefits case rested on roughly doubling the British Columbia intertie, thousands of megawatts, while what was actually announced amounted to around 150 megawatts of restoration.

That gap is not a failure of nerve, and it is certainly not a failure of talent; our regulators and system operators are full of capable people. It is that no institution is paid to think past its own border. Each province optimises within its own lines because that is the only mandate anyone has given it. Institutions optimise for what they are asked to optimise.

The wire is easy. Who plans it, who owns it, and who pays when the benefits land on one side of a border and the costs on the other: that is the hard part.

The governance is the project.

A new customer the size of the grid.

And now the newest and largest version of the same problem. Alberta’s data-centre connection queue is asking for 20.7 gigawatts. The province’s all-time record demand is 12.8 gigawatts. The amount actually cleared to connect before 2029 is 1.2 gigawatts.

Those numbers get offered as proof of a bubble, and they might be. But follow a single data centre trying to plug in, and you reach a question our market has no settled answer for.

When a customer arrives wanting to add more load than the system has ever carried, and may or may not still be there in a decade, who pays for the wires, the substations, the reinforced lines built to serve it?

Under traditional rate-base regulation, the answer is everyone. The utility upgrades the grid and recovers the cost from every ratepayer over decades — a feature of being a natural monopoly. If the demand proves real and durable, that is a fine bequest: future customers inherit a stronger grid. If it evaporates, citizens end up spending twenty years paying off a substation built for a server farm that never arrived, or that arrived and left.

Here the century argument cuts both ways, and it exposes the blind spot that matters most, because this is the one still being decided, and the sums are the largest on the table.

The principle that resolves it is old and unglamorous: cost causation. The customer who causes the cost should carry it.

A large, flexible, possibly-transient load should connect on cost-causative terms and pay a premium for its standing claim on everyone else’s grid, the right to lean on a system it did not build on the night its own power falls short. Ohio’s regulators have reached for a version of this with an eighty-five-per-cent take-or-pay condition on large loads; Virginia has drawn data centres into a separate rate class. These are early, imperfect attempts at exactly the right idea.

Price the connection correctly and the market does the sorting no amount of argument can. If a use of AI cannot cover the cost of the power it demands, that use retires itself, and the wires it paid to build remain for whatever comes next.

Because here is the part that should change how we see the whole boom. Whichever way the AI story ends, Canada is left holding much the same prize. If AI succeeds, efficiency compounds and today’s forecasts prove too high. If the optionality games unwind, the connection requests quietly expire. If the bubble bursts outright, the campuses go dark.

Walk any of those roads to its end and look around: the servers are gone, but the substations, the reinforced lines, the upgraded transformers remain, engineered for decades. The dot-com crash left the world an ocean of dark fibre, and the entire digital century moved in at a discount. The AI boom may be laying the grid’s version: quiet conductors, waiting for tenants.

But that inheritance is a gift only if the right people paid for it. Get the tariff wrong and the same wires become a stranded asset, socialised onto the households least able to carry it.

Why we keep getting it wrong

So why do these blind spots endure? Partly because pricing what a market has never priced is genuinely hard. But partly because we are not honest about cost in the first place.

Bent Flyvbjerg, who has studied more megaprojects than anyone alive, finds that fewer than half come in on budget, fewer than one in ten on budget and on time, and one in two hundred delivers on budget, on time, with the benefits promised. That consistency rules out mere bad luck.

He separates two causes: optimism bias, where estimators are honestly wrong, and strategic misrepresentation, where costs are deliberately lowballed to get a project past the point of no return, because once the hole is dug, nobody wants to fill it back in.

Seventy years of data should have cured honest optimism by now. It hasn’t, which points further toward the deliberate end of the spectrum than we like to admit.

A century-long denominator is not a licence to lowball the numerator. It is the opposite. It is the only argument strong enough to let a project face a fully honest price tag and still earn its approval.

The fix is a market, not a machine

None of this is solved by building more, or by building less. It is solved by fixing the rules that price and allocate. Four moves, in rough order of how overdue they are.

Price duration. Procure storage on terms that pay for multi-day reliability, not just the daily arbitrage cycle, so the thirty-seventh hour has a market before we need it.

Allocate cross-border cost. Stand up a body with an actual mandate to plan interties across provincial seams, and an explicit formula for who pays when benefits and costs fall on opposite sides of a line, with the federal government as financier and convenor rather than commander.

Make new load pay its share. Connect large flexible loads on cost-causative terms, with a backup-access premium for the standing claim they place on everyone else’s grid. This is the one that matters most right now, and the one our rules are least ready for.

And underneath all three, budget honestly. Independent scrutiny before commitment, and a published risk register, the discipline a serious private proponent already imposes on itself when it says, in effect, no long-term contract, no project.

There is a choice to make with the proceeds of that backup premium, and here I will admit a preference rather than claim a proof. I would send them to the households pushed into energy poverty by rising bills: let the largest new users of the grid help protect the smallest. Reasonable people will land elsewhere, and that is fine; it is a values position, not a market-design necessity. But the mechanism that makes the choice possible, charging new load for the backup it relies on, is not a values position at all. It is just honest pricing.

The thing we actually failed to build

We have spent a decade arguing about turbines and pipelines and panels, about which technology wins and what it costs.

Meanwhile the thing we have actually failed to build is none of those. It is a market that pays for duration, an institution that can plan across a border, and a tariff that charges a new customer for the grid he leans on. The country that builds those will quietly out-build the ones still arguing about the hardware. And in fifty years, the price will have become history, and the value will have become inheritance.

That was always the argument. It took me four tries to notice I was making it.