The Man Selling the Picks and Shovels Wants You to Relax About the Gold Rush
Jensen Huang has a number: half a million. That is how many AI-related jobs he says have been created in the past few years, and he has the floor at every conference where someone worries that AI will eat their career. He also has a theory. When steam engines made coal more efficient, Britain burned more coal, not less. The same Jevons paradox, he argues, applies to software: make coding cheaper, and demand for code rises. Fortune
Huang did not invent the jobs data he cites. He pulled it from his own remarks at the SCSP Memos to the President podcast, where he spent part of last week telling anyone listening that the AI job apocalypse is a fantasy conjured by people with what he called a "God complex." He was not naming Dario Amodei, exactly — but no one in the room needed the nameplate. Fortune
The problem is that Huang's financial interests and his labor-market optimism point in exactly the same direction. Nvidia's data center revenue — essentially its AI chip business — has climbed from roughly 39 percent of total sales in fiscal year 2022 to approximately 88 percent in fiscal year 2025. Gaming, once the core business, now accounts for about 7 percent. Nvidia earns when the entire economy rushes to build out AI infrastructure, regardless of whether that buildout ultimately creates or destroys net employment. That is not a conflict of interest that invalidates his argument. It is a structural incentive that deserves more scrutiny than it usually gets. Fortune
The empirical data is more ambiguous than either side admits. The half-million AI jobs figure comes from Huang's own remarks and lacks independent verification against Bureau of Labor Statistics data. The Indeed hiring data — showing software engineering demand rising — reflects current demand, not the market after widespread AI deployment. These are real data points, not invented ones. They are also selectively chosen, and presented without the hedging that independent economists typically apply to labor-market projections. Fortune
The Jevons paradox argument, borrowed from Apollo chief economist Torsten Slok, is intellectually coherent. Apollo Global Management — whose research Huang cited on the podcast — has documented historical precedent: making a resource cheaper tends to increase total consumption of it, not decrease it. If that mechanism applies to legal, consulting, and financial services, the employment picture from AI could look nothing like the mass-displacement scenario. Whether the analogy maps to knowledge work at scale is an open empirical question.
Amodei has not responded publicly. Anthropic declined to comment. His silence is notable: a CEO who has made frank discussion of AI risk part of his public identity chose not to reply to a direct, public challenge. Either the argument does not merit a reply, or a reply would require conceding ground that complicates Anthropic's position. Business Insider
The Indeed data Amodei's critics cite cuts both ways. Software engineering demand rising even as AI coding tools proliferate suggests AI may be increasing the demand for technical work — more code written means more code to review, integrate, secure, and maintain. That is consistent with the Jevons reading. It is also consistent with a labor market where the people being displaced are not the people posting job openings on Indeed. The net effect on entry-level white-collar employment remains genuinely uncertain, and neither the half-million jobs figure nor the 50-percent displacement estimate is solid enough to settle the question. Fortune
Huang's closing argument on the podcast was pragmatic. If the United States convinces a generation of college graduates that software engineering has no future, and demand for software keeps growing, that is a self-inflicted wound. He is not wrong about that. The trouble is he has the most structural reason in the room to want AI adoption accelerating — and the least personal incentive to flag the costs if it does not.