The AI jobs apocalypse may be farther away

Anthropic’s own research has weakened the case for an imminent AI jobs apocalypse. Its March analysis found “no systematic increase in unemployment for highly exposed workers since late 2022,” while deployment remains far below what current systems could theoretically achieve.

Claude covers 33% of tasks in the computer and math category, Anthropic reported, even though it could theoretically handle nearly 100%. Datacenter spending is surging, but productivity growth has not matched the sweeping predictions surrounding the technology. Labor productivity was slower during the first three years of the current AI era than during the information technology boom that began in the mid-1990s.

That gap has prompted a more cautious public debate. Anthropic co-founder Dario Amodei previously said AI could eliminate half of all entry-level jobs within one to five years and might become a “general labor substitute for humans.” But OpenAI CEO Sam Altman said in May that he no longer expects the kind of mass unemployment some AI companies have promoted.

Recommended reading

Nvidia commits $1 billion to South Korea’s AI buildout

“I don’t think we’re going to have the kind of jobs apocalypse that some of the companies in our space advocate or talk about.”

Why incomplete automation may preserve jobs

One argument against the most sweeping predictions is known as the O-ring theory, named after the inexpensive component whose failure caused the Space Shuttle Challenger disaster on 28 January 1986. The lesson is that a single task that cannot be performed reliably can constrain an entire system.

Applied to AI, the theory suggests that partial automation may increase the value of tasks machines cannot complete. AI could help highly skilled workers by taking over routine duties, or allow less-skilled workers to take on more responsibilities as systems handle expert tasks.

A recent study reached a similar conclusion: “despite strong substitution at the task level, overall employment effects are modest,” because productivity gains at firms adopting AI can offset falling demand in exposed occupations.

The evidence is still early. Jed Kolko has noted that research on AI’s labor-market effects remains in its infancy, and Amodei’s one-to-five-year forecast still has almost four years to run. The Federal Reserve says business adoption is expanding quickly. MIT economist David Autor also argues that AI continues to improve, with no obvious ceiling in sight.

“A lot of people have noticed that the world is not changing as fast as they predicted.”

AI’s cost and technical limits

The more cautious outlook does not eliminate the possibility of a disruptive future. Daron Acemoglu, the Nobel Prize-winning economist, said AI insiders remain convinced that artificial general intelligence is close. Elon Musk still predicts that “AI+Robots will be able to do everything, resulting in universal high income. Work will be optional.”

But technical and economic constraints are becoming harder to ignore. Autor says that “not everything is a computational problem”: AI can replicate language, yet it struggles to connect language with the physical world and continues to make critical errors.

The infrastructure bill could be enormous. The International Energy Agency expects datacenter electricity demand to more than double by 2030, reaching about 945 terawatt-hours—more than Japan’s total energy consumption. Some estimates suggest datacenters could eventually absorb 20%, 30% or 40% of GDP in investment.

That spending is also vulnerable to rapid depreciation as newer models replace systems built only months earlier. Acemoglu argues that AI model companies are “never going to make money” while losing hundreds of billions of dollars annually.

Public opposition is rising alongside those costs: seven in 10 Americans oppose building AI datacenters in their area, citing energy use and higher local electricity prices. The technology may yet transform work, but its promised future now faces a harder question than capability alone: whether society can afford it.