In September, somebody paid $3.4B for the right to make their own electricity.
The buyer has no name in the deal record. The record calls it a frontier AI lab, and then it stops. Reports since then have named a specific lab as the buyer. Neither company has confirmed those reports, and the unconfirmed name is not the point of the deal anyway. The point is what the money bought.
The scarce asset in AI is no longer the chip order. It is the power connection, and one buyer just paid $3.4B to own that connection outright.
What happened, in plain terms: OnePWR Solutions sold its power generation platform in September 2026 for $3.4B. The deal adviser’s own listing names the buyer only as a frontier AI lab. The sale is confirmed. Everything beyond the price, the date and the platform rests on reporting that the two companies have not endorsed, so this piece stays on the confirmed ground.
What exactly changed hands?
A power generation platform is the machinery of supply, a different asset class from a building with servers in it: generation equipment, the orders placed for more of it, and the work of putting that equipment into the ground and into service. The buyer did not rent capacity on somebody else’s schedule. The buyer took ownership of the supply itself.
That distinction matters because a data centre runs on two clocks. The first clock is construction, which builders know how to price; the second is the grid connection, which nobody prices well, because the queue belongs to somebody else. Buying the platform moves the second clock in-house.
Why buy the power instead of the power contract?
For two years, the argument in most boardrooms ran one way, and it was mostly right: chips decide who wins the AI race. A second argument has gained ground every quarter since then: the grid queue decides who waits.
Both beliefs describe a real constraint. They just describe different years. The constraint moved from the chip order to the connection date, and capital follows the constraint. A place in a queue is a promise held by a utility, and owned generation is an asset held by the firm. The $3.4B is what that difference costs at the frontier end of the market.
What does Alberta charge for each version of this move?
Alberta wrote its answer into law before this deal closed. The province’s data centre regulation has been in force since June 2026, and it directs the grid operator to put projects first when they arrive paired with their own new generation or storage. In Alberta, bring your own power is the sorting rule for the queue.
The queue needs a sorting rule: requests on the grid operator’s connection list stood at about 19,565 MW on July 30, 2026. The first phase of the province’s allocation allowed 1,200 MW, and that room is fully taken. Every project behind it competes on the strength of what it brings with it.
Then there is the price ladder. From December 31, 2026, a grid-connected data centre at 75 MW or more pays a 2% levy on its computer hardware, fully creditable against Alberta corporate income tax. A site that generates its own power pays 1%. A fully off-grid site pays 0%. The ladder is set out in Alberta’s data centre levy can drop to 0%. The host town can’t tax the servers, and the province’s own data centre page carries the regulation and the queue in one place.
Read the ladder as a set of instructions. Grid power costs the most. Your own power costs half. Total independence costs nothing. Alberta did not plead with labs to bring generation, and it did not ban them from the grid either. It priced each choice and let the spreadsheet decide. The OnePWR buyer made the same calculation from the other side of the border.
The municipal layer still runs on its own clock, as Calgary’s data centre pause hangs on one undefined word shows. A province can sort its queue and price its levy, and a project still waits on a town’s definition.
Who is already building the Canadian version?
Meta is, in Sturgeon County, where its campus is under construction at 1 GW and more than $13B, the company’s first data centre in Canada. The build carries its own funded generation and grid infrastructure with it. That is the OnePWR logic at campus scale, with the power plan built into the project itself.
Nova Scotia’s largest build runs on a related promise. The $10 billion Nova Scotia data centre buys 215 jobs and a promise to unplug first tells that story: a campus that draws on the provincial grid first and brings its own supply behind it. The sequence differs by province. The direction holds.
What should a Canadian owner take from a deal with no name on it?
Your firm will never buy a power platform, and it does not need to. It will sign contracts with firms whose costs now include power they had to secure the hard way. When a provider’s price rises, the honest question is which clock drove it: the build, or the connection.
Ask two questions at your next renewal. Where does the power for your workload actually come from, in which province, under which queue? And who carries the delay when the connection date moves? A vendor who cannot answer the first question plainly has not priced the second one either.
The recap, in three lines. Alberta had already sorted its queue around exactly that risk, and priced the exits. The Canadian builds that move fastest are the ones that arrive with power in hand. The checked figure sits above both: $3.4B, paid by a frontier lab for generation because the queue had become the risk.
Follow AI Magazine Canada for the next move. The tracker behind this piece updates every time a province, a town or a lab changes the picture.
