Canada's Data Centre Boom Is Missing Its Electricians

The Economist told its readers this month that artificial intelligence has already created roughly a million jobs in America, and that the long-promised jobs apocalypse has been postponed. The post carrying that claim has been seen close to a million times.

Most of those jobs are not software jobs. They are electrical and civil construction jobs, because a data centre is, in employment terms, an enormous power and building project with a small permanent staff bolted on at the end. Meta says as much in its own announcement of its first Canadian facility: over 3,000 construction workers at peak, and more than 300 operational jobs. Ten to one.

Which raises the obvious question for readers on this side of the border. Canada has spent two years announcing gigawatts. Is any of it showing up in Canadian payrolls yet?

We pulled both countries' payroll surveys, matched the NAICS codes line for line, and compared the identical four-year window. The answer is no, and one line in particular runs the wrong way.

The trade that builds the connection is shrinking

NAICS	Industry	United States	Canada
2371	Utility system construction	+13.6%	−11.8%
237	Heavy & civil engineering	+12.1%	−2.1%
22	Utilities	+9.1%	+7.5%
23	Construction, all	+7.2%	+5.6%
5415	Computer systems design	−3.7%	+1.4%
—	Total employment	+4.3%	+4.8%

Start at the bottom of that table, because it is what makes the top of it legible. Total employment grew at almost exactly the same rate in both countries over these four years, 4.3 percent in the United States and 4.8 percent in Canada. There is no macro story here, no divergent business cycle to explain away the rest of the column. The two labour markets grew in step.

Now read the top line. US utility system construction employment rose 13.6 percent between June 2022 and June 2026, adding 76,300 jobs. Canada's fell 11.8 percent, losing 9,284. That is a 25-point swing in the single trade whose job is to build substations, transmission connections and the electrical distribution a data centre cannot open without.

Heavy and civil engineering construction, the tier above it, tells the same story more quietly: up 12.1 percent in the US, down 2.1 percent in Canada.

Everything else in the table is unremarkable. Utilities employment grew in both countries at a similar clip. Total construction grew in both. Computer systems design, the white-collar software services code, was slightly down in the US and slightly up in Canada, which is its own interesting finding and not this one. The divergence is concentrated, almost surgically, in the two codes that do heavy electrical infrastructure.

The computer systems design line is worth a sentence on its own, because it cuts against the usual Canadian anxiety. That code covers IT consultancies, systems integrators and much of the country's software services employment, and it is the part of the economy most people assume AI is hollowing out. Over these four years it shrank 3.7 percent in the United States and grew 1.4 percent in Canada. Whatever is happening to American software services employment, Canada is not experiencing a worse version of it. The Canadian shortfall is not in the knowledge work. It is in the concrete and the copper.

Method: US figures are Bureau of Labor Statistics Current Employment Statistics, seasonally adjusted, read from the BLS public API. Canadian figures are Statistics Canada's Survey of Employment, Payrolls and Hours, table 14-10-0201-01, all employees, which is not seasonally adjusted, so we used a same-month June-to-June comparison on both sides. That removes seasonality without needing an adjustment, and it constrains the window to June 2026, the latest SEPH month. Both are establishment surveys of employers, which is what makes them comparable; both use the same NAICS codes at the same level of detail.

One caution we will state rather than bury. Canada's Labour Force Survey, the household survey most Canadian coverage quotes, shows utilities employment essentially flat over a similar window, up about 1,000 jobs. SEPH shows it up 7.5 percent. For an industry with roughly 140,000 payroll employees, the household survey's sample is too thin to settle a question like this, and we have used the payroll survey throughout for that reason. Where they disagree on small industries, SEPH is the better instrument.

Why this is the number that matters

Everything else about a data centre can be bought or shipped. Land, steel, cooling, chips, even the operating staff. The one thing that cannot be procured on a schedule is a grid connection, and building one is exactly what NAICS 2371 does.

The United States is currently hiring that trade at more than three times the rate of its overall labour market. Canada is shedding it. That is not a story about who wants data centres. It is a story about who has started building the electrical infrastructure to hold them.

And Canada plainly wants them. Ottawa committed CAD 2 billion to domestic AI compute capacity in December 2024. Alberta has set a target of attracting $100 billion in data centre investment by 2030. A Cohere and CoreWeave facility in Toronto, backed by C$240 million in federal money, is the largest AI installation in the country. Analysts tracking announced capacity put Canada second in the world behind the United States, at roughly 13 GW against 104 GW, which sounds like a race until you notice that the Canadian figure is overwhelmingly announced rather than operating. BCG's assessment of Canadian data centre capacity puts what is live today closer to 0.3 GW, against roughly 1.3 GW needed by 2030 to serve Canadian workloads domestically.

The Meta project makes the timing legible. Meta broke ground in Sturgeon County, north of Edmonton, in July 2026, on a C$13 billion, 1 GW campus across 1,750 acres. The power for it comes from Pembina's Greenlight Electricity Centre, a new gas-fired generating station in the same county, and that station is scheduled to enter service in late 2030. The construction employment for a project like this does not arrive on the day of the announcement. It arrives when the electrical work does.

So the honest reading of the table is not that Canada has lost. It is that Canada is currently in the gap between the announcement and the shovel, and the payroll data has not started to move because there is not yet very much to build. The −11.8 percent is what the pre-build period looks like in a statistic.

The part that should worry planners

The gap is not free, for two reasons.

The first is that the trade shrinks while it waits. A utility contractor that loses 9,284 positions over four years does not hold those crews in reserve. Apprenticeships lapse, journeypersons move to other provinces or other industries, and the firms that would have bid the work scale down their capacity to bid it. When the Sturgeon County electrical scope and everything queued behind it lands at once around 2029 and 2030, it lands on a workforce that has been contracting for the entire preceding decade-half. The likely result is not that the projects fail. It is that they get more expensive and slower, and that a meaningful share of the skilled labour is imported from exactly the American market that spent this period expanding.

The second is that the sequencing is visible to the people deciding where the next campus goes. A hyperscaler choosing between jurisdictions is pricing time-to-power, and time-to-power is a function of whether the local electrical construction industry has spare capacity. The US number and the Canadian number are inputs into that decision, not just descriptions of it.

None of this is an argument that Canada should have built speculative substations in 2023. It is an argument that the interval between committing to an industrial strategy and staffing the trades it depends on is the part that gets skipped, and that it is now measurable.

What to watch, and when

This is a falsifiable claim, which is the useful kind.

If the announced Canadian build-out is real and on schedule, NAICS 2371 employment in the SEPH data should stop falling and turn positive during 2027, roughly two to three years ahead of the 2030 in-service dates, because heavy electrical work leads energisation by about that much. Watch the June figure each year; the same-month comparison used here is reproducible in a few minutes from the public table.

If it is still negative in June 2028, the announcements and the construction schedule have decoupled, and the reasonable conclusion is that Canada is hosting the press conferences for a build-out that is being staffed somewhere else.

The gigawatt numbers will keep being announced either way. The electricians are the ones who will tell you whether it is happening.