Skip to content
Welcome, thoughtbot's Giant Robots listeners!Start here
Data Centres: Boom and Backlash
22OCT

Canada finds the hiring door narrowing

3 min read
16:40UTC

The Bank of Canada set Canadians in fully AI-exposed occupations against workers facing no exposure at all. The gap in how readily the two groups find work widened by 11.7 percentage points; the gap in how often they lose it moved 0.1.

IndustryDeveloping
Key takeaway

Canada's data shows AI narrowing the hiring door, not pushing incumbents out through the exit.

The Bank of Canada published a staff study on 24 August 2026 using Statistics Canada Labour Force Survey microdata, comparing how readily workers in fully AI-exposed occupations find jobs against workers in occupations scored at zero exposure⁠1. Across the 2015 to 2019 average, the gap between the two groups stood at 2.2 percentage points. In 2025 it stood at 13.9 points.

Read that number carefully, because it invites a false reading. No group's job-finding rate fell by 13.9 points. The distance between exposed and unexposed workers widened by 11.7 points, and 13.9 is where that distance now sits. The separation-rate gap, which measures the difference in how often each group loses work, moved from 0.0 points to 0.1. Dismissals did not do this. The unemployment-risk gap widened more modestly, from 1.9 to 2.8 points, which is what you would expect of workers who are not losing jobs but are queueing longer to get one.

A differential survives an objection a level cannot. A general downturn depresses hiring for exposed and unexposed workers alike and leaves the distance between them where it was. For a graduate applying into an exposed occupation this autumn, the whole effect sits in a vacancy that never opens, and no notice period, severance payment or redundancy tracker records its absence.

One institution reads the same period the other way. Pablo Hernandez de Cos, General Manager of the Bank for International Settlements, the Basel institution that acts as a bank for central banks, told the Global Fintech Fest in Mumbai on 10 September 2026 that actual displacement remains limited and that firms sit in wait-and-see mode, even while nearly 80% of firms discuss plans to automate production and substitute labour⁠2. Federal Reserve governor Michael Barr found little evidence of economy-wide displacement in July, and governor Lisa Cook said that month that the dire predictions had not arrived. A New York Fed survey of service firms had already sketched the Canadian shape on 5 August, with adoption climbing, AI-driven redundancies rare, and hiring suppression falling on college-educated applicants.

Deep Analysis

In plain English

Two different numbers are moving in this study, and mixing them up changes the story completely. The "separation rate" is how likely you are to lose the job you already have. That has barely budged: 0.0 to 0.1 percentage points. If you already work in an AI-exposed occupation, you are not visibly more likely to be let go than you were a decade ago. The "job-finding rate" is how likely someone without a job is to land one. That gap has widened from 2.2 to 13.9 percentage points between AI-exposed and unaffected occupations. If you are trying to get into one of those fields from outside, the door has become much harder to open. So the honest summary is not "AI is firing people". It is closer to "AI is making it harder to get hired into some jobs, while doing comparatively little to the people already in them".

Deep Analysis
Root Causes

Firing an employee already on the payroll carries severance costs, morale costs among the staff who remain and, in many jurisdictions, legal exposure. Leaving an open role unfilled carries none of those costs. When AI lifts output per worker, a firm's cheapest first move is to stop backfilling a departure or delay a planned hire, not to dismiss someone already doing the job.

That asymmetry predicts exactly the shape the Bank of Canada found: a job-finding rate that falls sharply while the separation rate barely moves, because the two adjustments sit on opposite sides of a very different cost calculation, and firms will exhaust the cheap option first.

What could happen next?
  • Meaning

    The debate over AI's labour effect should centre on hiring difficulty for outsiders, not headline job losses among incumbents, because the two respond to entirely different cost structures.

  • Risk

    New graduates and career-changers in AI-exposed fields face a disproportionate share of the adjustment, even while overall unemployment statistics stay calm.

First Reported In

Update #20 · AI closes the hiring door, not the exit

Bank of Canada· 21 Sept 2026
Read original →
Different Perspectives
Mohsen Rezaei
Mohsen Rezaei
The senior Iranian security official and former IRGC commander told Press TV that US-linked technology infrastructure in the region could become a target if war resumes. He named the UAE and valued the assets at more than US$2 trillion, without a breakdown.
Mykhailo Shelemba
Mykhailo Shelemba
Datagroup's chief executive says the Kyiv data centre hit on the night of 18 to 19 October is beyond repair. Services continue from backup sites, and no staff were hurt.
Kristen Umstattd
Kristen Umstattd
The Loudoun supervisor voted against fast-tracking Amazon's rezoning of the Ashburn campus. She said there were no figures yet on homes, school or road costs.
Akishima residents' litigation group
Akishima residents' litigation group
The group formed on 18 October and plans to sue Nihon GLP by year-end to stop eight data-centre buildings. Its lawyers will argue rights to life, health, clean water and a natural environment after mediation failed in August.
Óscar López
Óscar López
Spain's Digital Transformation minister said on 22 October he could not confirm when the data-centre decree will be approved. He added that Google has always committed to investing in Spain.
Google
Google
Google said it would comply with LVV's order to stop preparatory works at Muhos and Kajaani until environmental assessments finish. A spokesperson conceded the company had fallen short.

Written with AI from Lowdown's published reporting. Not individually reviewed by an editor. Editorial standards.