Key Takeaways

  • Paul Krugman argues that AI infrastructure spending can weaken job creation elsewhere, even without mass automation-driven unemployment.
  • Research points to slower hiring and fewer entry-level opportunities, while economy-wide employment effects remain limited.
  • Business leaders face a broader capital-allocation question as billions flow into data centers, chips, power, and AI software.

Paul Krugman has sharpened the economic debate around artificial intelligence by focusing less on robots replacing entire professions and more on what today's AI investment boom may be displacing.

His October 2026 Substack argument centers on "crowding out." Capital, skilled labor, electricity, construction capacity, and management attention committed to AI infrastructure are resources that cannot simultaneously be deployed elsewhere. Krugman's illustrative comparison is striking: putting $8 billion, roughly the cost of a data center, into a non-AI sector would support more than 10,000 jobs on average (source).

That does not mean the same investment in an AI data center creates no employment. Such projects require construction workers, electrical engineers, technicians, security personnel, networking specialists, and ongoing operations teams. The issue is job intensity. Data centers can absorb vast sums while creating relatively few permanent positions compared with labor-intensive investments in manufacturing, healthcare, hospitality, housing, or infrastructure.

Mass technological unemployment does not need to arrive for the labor market to feel the effects.

A 2026 study from the National Bureau of Economic Research found that nearly 70% of firms across the U.S., U.K., Germany, and Australia report using AI. Executives expect the technology to raise productivity by 1.4% and reduce employment by 0.7% over the next three years, largely through slower hiring rather than sweeping layoffs.

That distinction matters to enterprise leaders. A company that keeps its existing workforce but recruits fewer analysts, developers, support agents, or junior marketers is still changing the employment landscape. The adjustment happens quietly, one unfilled vacancy or smaller graduate intake at a time.

Evidence from the Federal Reserve Bank of Richmond reinforces the mixed picture. Executives reported 1.8% AI-associated labor-productivity growth in 2025, while average employment effects connected to AI were negligible. Large firms, however, expect employment to fall 0.8% in 2026.

So far, this looks more like task reallocation and hiring restraint than wholesale job destruction. OpenAI, Microsoft Copilot, and Anthropic Claude increasingly handle first drafts, code suggestions, document summaries, customer-service responses, and routine analysis. Experienced employees can use those systems to increase output. But what happens to the junior worker who previously performed those first-pass tasks?

That question is becoming harder to dismiss. Research associated with Stanford University's Digital Economy Lab found that workers aged 22 to 25 in highly AI-exposed occupations experienced a 16% relative employment decline compared with more experienced workers (source). The finding does not establish an economy-wide collapse, but it suggests entry-level pathways may be narrowing in selected occupations.

Automating entry-level roles presents a direct operational risk. Junior work is not only low-cost production. It is how employees acquire judgment, institutional knowledge, and the practical experience needed for senior roles. Companies that automate too much foundational work could discover later that their leadership and specialist pipelines have thinned.

Capital allocation adds another layer. AI spending requires advanced chips, grid connections, cooling systems, land, and financing. When demand for those inputs surges, costs can rise for other industries. That is the crowding-out mechanism Krugman is highlighting, and it extends beyond direct employment counts.

The policy response need not assume AI adoption should stop. The OECD AI Principles, adopted in 2019 and updated in 2024, emphasize transparency, human oversight, and responsible deployment. The NIST AI Risk Management Framework, released in 2023, similarly gives organizations a structure for evaluating AI risks, including consequences for people and business processes.

For executives, that can translate into tracking more than cost savings. Hiring rates, junior-role availability, training hours, task changes, error escalation, and workforce mobility can reveal whether an AI program is building durable productivity or merely removing the first rung of the career ladder.

The current evidence remains nuanced. AI has not produced economy-wide job collapse, and its productivity gains are becoming measurable. Still, Krugman's warning shifts the conversation in a useful direction: the economic cost of AI may appear not only in jobs eliminated, but also in investments deferred, positions never opened, and careers that struggle to get started.