AI Displacement Immunity

AI displacement immunity describes how protected a business is from being copied, automated away, or made obsolete by artificial intelligence. A company scores high when its earnings rest on physical assets and real-world operations, like mining, freight, or owned real estate, that a model cannot reproduce no matter how capable it gets.

What is AI displacement immunity?

The idea sorts companies by a single test: if AI got much better tomorrow, would this business be in trouble? A stock-photo library or a basic copywriting service scores low, because a model can do that work. A copper miner, a freight railroad, or a billboard owner scores high: the metal still has to be dug, the container still has to move, the billboard still occupies real estate.

Academic work maps the same fault line, and it runs along the divide between information work and physical work. The peer-reviewed “GPTs are GPTs” study quantified the exposure of every U.S. occupation to large language models:

Our findings reveal that around 80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of LLMs, while approximately 19% of workers may see at least 50% of their tasks impacted.

— Eloundou, Manning, Mishkin & Rock, Science (2024)

The same paper found the highest-exposure jobs were higher-paid information roles such as writers, accountants, and tax preparers, while the lowest-exposure occupations were hands-on work like maintenance and construction (Eloundou et al., Science 2024). The International Monetary Fund reached a parallel conclusion: about 40% of jobs worldwide, and roughly 60% in advanced economies, are exposed to AI, because rich countries lean on cognitive, office-based work a model can shoulder (IMF, 2024). Read the other way, the cash flows AI struggles to touch are the ones rooted in the physical world.

Where does the term come from?

The phrase grafts an investing question onto a labor-economics concept. “Displacement” is the economists’ term for technology taking over tasks people used to perform, formalized by Daron Acemoglu and Pascual Restrepo in their task-based model of automation:

Automation, which enables capital to replace labor in tasks it was previously engaged in, shifts the task content of production against labor because of a displacement effect.

— Daron Acemoglu & Pascual Restrepo, Journal of Economic Perspectives (2019)

“Immunity” flips that lens from workers to cash flows: the screen asks which businesses sit in tasks AI cannot perform, so their earnings escape the displacement effect. The label comes from thematic investing rather than the journals, but the mechanism it names is the one Acemoglu and Restrepo modeled.

How is AI displacement immunity measured?

There is no single standard score, so practitioners invert the academic exposure measures. The AI Occupational Exposure (AIOE) index of Felten, Raj, and Seamans scores every U.S. occupation by linking AI progress in fields like image recognition and language modeling to the abilities each job relies on (Felten et al., Strategic Management Journal 2021); Eloundou and coauthors did the same at the task level for LLMs. Applied to a company, the same logic maps revenue to the underlying work: how much is information a model can absorb, and how much is bound to physical operations.

That exposure is no longer hypothetical. Tracking ADP payroll records, Stanford’s Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen found early-career workers aged 22 to 25 in the most AI-exposed occupations saw a 16% relative employment decline after late 2022, with “employment declines… concentrated in occupations where AI is more likely to automate, rather than augment, human labor” (Brynjolfsson, Chandar & Chen, Stanford Digital Economy Lab, 2025).

How is AI displacement immunity used in thematic investing?

It is the central screen behind the Heavy Asset Low Obsolescence (HALO) concept. Where physical asset intensity measures how tangible a company is, AI displacement immunity asks whether those tangible operations are the kind AI cannot take over. A data-entry firm and an iron-ore miner can both own buildings, but only one of them does work a model can replicate.

Why does AI displacement immunity matter for investors?

The screen separates two risks that markets often price together. A software vendor and a port operator may both look like steady cash generators, yet a sharp jump in model capability threatens one and barely grazes the other. Weighting businesses whose moat is a physical asset or a real-world operation aims to hold earnings that survive whatever the next model release can do. It is the opposite bet to owning the AI tools themselves, and it leans on the most durable finding in the labor research: exposure tracks how much of a job is reproducible information, and physical work sits at the protected end (Eloundou et al., Science 2024).

Related terms & concepts

FAQ

How do you measure AI displacement immunity?

There is no single number. Analysts borrow academic exposure measures, such as the AI Occupational Exposure (AIOE) index of Felten, Raj, and Seamans (Strategic Management Journal, 2021) or the task-level LLM rubric of Eloundou et al., then invert them: they judge how much of a company's revenue depends on physical assets and real-world operations versus information work a model can reproduce.

Which kinds of companies have high AI displacement immunity?

Businesses built on physical assets and hands-on work, such as miners, freight and rail operators, utilities, and owners of physical real estate, tend to score high, while software, content, and routine information-processing businesses score low.

Where does the term AI displacement immunity come from?

It builds on the "displacement effect" that economists Daron Acemoglu and Pascual Restrepo formalized in their task-based model of automation, where capital replaces labor in tasks it previously performed (Journal of Economic Perspectives, 2019). The investing term flips that lens from workers to cash flows, asking which businesses sit in tasks AI cannot take over.

Is there evidence that AI displacement is already happening?

Yes. Tracking ADP payroll records, Stanford researchers Brynjolfsson, Chandar, and Chen found early-career workers aged 22 to 25 in the most AI-exposed occupations experienced a 16% relative decline in employment after late 2022, concentrated where AI automates rather than augments work (Stanford Digital Economy Lab, 2025).

Sources & references

  1. Akros Thematic Index Methodology Framework (EN) · Akros Technologies, Inc., 2025-11-01
  2. GPTs are GPTs: Labor market impact potential of LLMs · Eloundou, Manning, Mishkin & Rock · Science, 2024-06-21
  3. Gen-AI: Artificial Intelligence and the Future of Work · Cazzaniga et al. · International Monetary Fund (Staff Discussion Note SDN/2024/001), 2024-01-14
  4. Automation and New Tasks: How Technology Displaces and Reinstates Labor · Acemoglu & Restrepo · Journal of Economic Perspectives, 2019-05-01
  5. Occupational, Industry, and Geographic Exposure to Artificial Intelligence: A Novel Dataset and Its Potential Uses · Felten, Raj & Seamans · Strategic Management Journal, 2021-12-01
  6. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence · Brynjolfsson, Chandar & Chen · Stanford Digital Economy Lab, 2025-11-13