AI spending boom collides with sluggish productivity
Anthropic’s latest employment analysis reveals that AI deployment lags behind theoretical capabilities, raising sharp questions about the return on massive technology investments.
Anthropic published an analysis in March showing that artificial intelligence is falling far short of its promised economic impact. The company’s Claude chatbot currently handles just 33% of tasks in the computer and mathematics category, despite theoretical capabilities nearing 100%. Crucially, the report found “no systematic increase in unemployment for highly exposed workers since late 2022.”
This represents a stark shift from the rhetoric of Anthropic’s own leadership. Co-founder Dario Amodei previously warned that AI could eliminate half of entry-level jobs within five years and become a general human substitute. Yet even OpenAI’s Sam Altman has softened his tone. “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,” Altman said in May.
For investors, the central problem is that soaring capital expenditure is not yet translating into economic output. Datacenter spending is accelerating rapidly, but labor productivity growth during the first three years of the AI era is actually slower than it was during the mid-1990s information technology boom. The market is beginning to price in this disconnect, with the tech-heavy Nasdaq falling roughly 8% since its early June peak.
The return on investment question
Economists point to the "O-ring" effect to explain why automation is not displacing workers as expected. Because AI cannot yet perform every task flawlessly, its deployment often increases the value of the human tasks it cannot replace. One study noted that reduced demand in exposed roles is offset by “productivity-driven increases in labor demand at AI-adopting firms.”
The deeper financial risk lies in the economics of the infrastructure buildout itself. Models depreciate rapidly as new iterations render older ones obsolete. “They are never going to make money,” said Nobel prize-winning economist Daron Acemoglu of AI developers. “They are losing hundreds of billions of dollars every year.”
The physical constraints are equally daunting. The International Energy Agency estimates datacenter power demand will more than double by 2030 to 945 terawatt-hours, exceeding Japan’s entire energy consumption. This insatiable appetite for power is stoking political backlash, with seven in ten Americans opposing local datacenter construction due to rising electricity costs.
The technology will undoubtedly continue to improve, and insiders remain confident that artificial general intelligence is imminent. However, as MIT economist David Autor noted, “Not everything is a computational problem.” The immediate threat to markets is not a dystopian takeover of human labor, but a capital expenditure cycle that may fail to deliver viable returns at a price the broader economy can sustain.
