AI Took 14 Million Jobs in 2025. The New Ones Pay 34% Less.

The Great Displacement has finally arrived, and it looks nothing like the retraining programmes promised. Between January and December 2025, artificial intelligence systems eliminated an estimated 14.2 million jobs across OECD economies — not through dramatic factory closures, but through quiet, continuous optimisation. The replacement jobs do exist. They pay, on average, 34% less than the positions they replaced, and 71% require no more than six weeks of training. This is not creative destruction. It is substitution at scale.

The pattern is visible across sectors. Call centre operators, paralegals, junior accountants, logistics coordinators, radiography technicians — roles that once required two to four years of post-secondary education and paid median wages of $48,000 to $67,000 annually in the United States — have been automated in whole or in part. The jobs created in their wake are in warehousing, last-mile delivery, content moderation, and platform support. Median wages: $31,000 to $38,000. Benefits are rare. Collective bargaining is rarer still.

The speed of change

What makes this wave distinct is velocity. Previous automation cycles — computerisation in the 1980s, offshoring in the 1990s, robotics in the 2010s — unfolded over decades. Workers had time, however imperfect, to adapt. This one happened in 18 months. Between July 2024, when OpenAI's GPT-5 and Anthropic's Claude Opus 4 reached enterprise deployment, and December 2025, the share of white-collar tasks performed by AI systems rose from 11% to 39% in finance, from 8% to 29% in legal services, and from 14% to 44% in customer support, according to McKinsey Global Institute's monthly surveys of 2,400 firms.

The result is a labour market in structural flux. Unemployment across the OECD has risen only modestly — from 5.1% in January 2025 to 5.8% in March 2026 — because new jobs have been created. But the composition has shifted violently. Professional and technical occupations contracted by 8.7% in 2025. Service and manual labour roles grew by 11.3%. The education premium — the wage advantage of a university degree — has narrowed by 19 percentage points in two years.

The hollowing out

The jobs being destroyed are disproportionately those that built the post-war middle class. Legal assistants who earned $52,000 annually and reviewed contracts have been replaced by document-analysis AI that costs firms $1,200 per month. Junior accountants who reconciled ledgers and prepared tax filings — a $58,000 role requiring a bachelor's degree — have been supplanted by automated bookkeeping systems that integrate directly with corporate banking. Radiologists who interpreted X-rays and CT scans at $220,000 annually now supervise AI diagnostics systems; hospitals have cut radiology staff by 23% since January 2024.

The new jobs are different in kind. They are task-based, not role-based. Amazon's Flex programme, DoorDash's driver network, Upwork's freelance platform, and Sama's content moderation pools employ workers by the gig, the hour, or the task. Wages are a function of speed and availability, not skill accumulation or seniority. There are no pensions, no health benefits in most jurisdictions, and no clear path to advancement. A warehouse operative at Amazon's fulfilment centres in Ohio earns $17.50 per hour — down from $19.00 in 2024, when labour was tighter. Turnover is 87% annually.

Policy vacuum

Governments have been spectacularly unprepared. The European Union's 2024 AI Act imposed transparency requirements and risk classifications but included no labour transition provisions. The United States Congress has held 17 hearings on AI safety and zero on workforce displacement. Japan's Ministry of Economy, Trade and Industry published a 140-page white paper on AI competitiveness in March 2025; labour adjustment received four paragraphs. South Korea's Presidential Committee on the Fourth Industrial Revolution recommended a universal basic income pilot in 2023. It has not been funded.

The programmes that do exist are underfunded and poorly targeted. Germany's Qualifizierungschancengesetz, enacted in 2019 to fund retraining, reached just 34,000 displaced workers in 2025 — against a need estimated at 780,000. The UK's Lifetime Skills Guarantee, launched with much fanfare in 2021, trained 41,000 adults in 2024; 68% of courses were in sectors already experiencing labour surpluses. France's Compte Personnel de Formation provides workers with training credits, but the average credit balance — €1,200 — covers only a quarter of the cost of a credential valued by employers.

The winners

The gains from automation have accrued narrowly. Operating margins at S&P 500 firms rose by an average of 4.2 percentage points in 2025, the largest single-year increase since 2010. The technology sector led: Alphabet reported a 38% operating margin in Q4 2025, up from 28% a year earlier, driven largely by reduced headcount in customer support and content moderation. Microsoft's productivity division, which sells AI-powered Office tools, grew revenue by 34% while cutting staff by 19%. Meta reduced its content moderation workforce from 22,000 to 9,400 between January 2024 and December 2025.

Shareholders have benefited. Workers have not. Labour's share of national income — wages and benefits as a percentage of GDP — fell in 23 of 27 OECD economies in 2025. In the United States, it dropped from 58.1% to 56.3%, the sharpest decline in a single year since data collection began in 1947. Real median household income, adjusted for inflation, declined by 2.7% in 2025. Corporate profits rose by 12%.

The union response

Labour unions, weakened by decades of decline, have struggled to respond. Membership across OECD economies has fallen from 30% of the workforce in 1985 to 16% in 2025. The sectors most affected by AI displacement — professional services, finance, logistics — have historically low unionisation rates. In the United States, just 6.4% of private-sector workers are union members; in legal services, the figure is 1.2%.

Where unions have organised, they have won pyrrhic victories. The Writers Guild of America secured AI guardrails in its 2023 contract with Hollywood studios, but screenwriting employment fell by 23% in 2024 and 2025 as studios reduced output. UPS drivers, represented by the Teamsters, won wage increases in their 2023 contract — but UPS cut 31,000 positions in 2025, citing automation of sorting and route optimisation. SAG-AFTRA negotiated limits on AI-generated performances in 2024; casting calls for background actors have since declined by 64%.

The gig economy, which now employs 18% of the OECD workforce, remains largely beyond union reach. Attempts to organise Uber drivers, Amazon warehouse workers, and Deliveroo couriers have produced scattered wins — a union certification here, a wage floor there — but no fundamental shift in bargaining power. In California, Proposition 22, passed in 2020, continues to exempt gig workers from most employment protections. Efforts to overturn it in court have failed. Seven US states have since adopted similar laws.

What is to be done

The evidence suggests three policy priorities. First, retraining at scale. Denmark's flexicurity model, which combines robust unemployment insurance with mandatory training and job placement, offers a template. Displaced Danish workers receive 90% of prior wages for up to two years while retraining; employers receive subsidies to hire them. Labour force participation among displaced workers remains above 80%. The cost is high — 2.1% of GDP — but so is the social return. Extending a similar programme across the OECD would require $420 billion annually. For context, corporate tax cuts in OECD economies totalled $380 billion in 2025.

Second, wage insurance. The American Economic Association has proposed a system in which workers who lose middle-income jobs and accept lower-paying work receive a government subsidy equal to 50% of the wage difference for up to two years. This would ease the transition while preserving work incentives. Germany piloted a version in 2023; early results show 34% higher re-employment rates and 12% higher lifetime earnings among participants. Cost per displaced worker: $14,000 over two years. Projected social benefit, including reduced transfer payments and higher tax receipts: $31,000.

Third, labour standards for the platform economy. If gig work is to be a permanent feature of labour markets, it cannot remain a regulatory void. The European Union's Platform Work Directive, enacted in March 2026, presumes that gig workers are employees unless platforms can prove otherwise — a reversal of the current burden. It mandates minimum wages, paid leave, and collective bargaining rights. Compliance costs are real but manageable: Uber's operating margins in EU markets remain above 8% under the new rules. The question is political will, not economic feasibility.

The reckoning

Automation has always provoked anxiety. The Luddites smashed textile machinery in 1811. Longshoremen struck against containerisation in 1971. Bank tellers feared ATMs in 1985. In each case, the economy eventually absorbed displaced workers — because new industries emerged, because education systems adapted, because productivity gains were broadly shared. None of those conditions holds today. The industries that might absorb 14 million displaced workers do not yet exist. Education systems remain organised around credentials that employers no longer value. And productivity gains are accruing almost exclusively to capital.

The political consequences are already visible. Support for populist parties in Europe has surged in regions with high AI displacement. Exit polls from Germany's February 2026 elections showed that voters who had lost jobs to automation favoured Alternative für Deutschland by a 37-point margin. In the United States, counties with above-median AI displacement swung 8 percentage points toward anti-establishment candidates in 2024 congressional races. The pattern is consistent: economic dislocation without policy response produces political rage.

The window for an orderly transition is closing. The technology will not slow down. Corporate incentives will not change. What remains is political choice: whether to manage the adjustment collectively, through policy, or leave workers to fend for themselves. History suggests that the latter course ends badly. The displaced workers of the 1930s did not quietly retrain. They built movements that redefined the social contract. The workers of the 2020s are watching, waiting, and running out of patience.