Steve Hanke is betting against the most popular AI storyline: widespread job replacement. In an email to Business Insider, the Johns Hopkins applied economics professor argued AI is not a near-costless upgrade to productivity, calling the belief that it will be “free to use and virtually costless to provide” delusional and “dumb.”

Hanke’s core claim is straightforward and operational. AI is “incredibly costly,” he said, and it is “very resource intensive.” He pointed to the physical requirements behind modern AI systems: “huge amounts of water, power, and physical capital” like graphics chips. In his framing, the practical constraint is not models getting smarter, it is whether companies can afford the scarce inputs required to run and scale them.

This is where the job story changes from vibes to balance sheets. Hanke told Business Insider that businesses will not simply fire everyone and swap in AI, because in many cases replacing humans could be more expensive than employing humans. That flips a common assumption that AI will automatically win on cost per output. Instead, Hanke says the economics start with infrastructure and energy, then move to labor substitution decisions. If your “compute bill” grows faster than your savings, the rational move is not mass replacement. It is selective automation where ROI beats payroll.

The resource constraint is not theoretical. Business Insider reports that Microsoft, Alphabet, Amazon, and Meta have projected roughly $700 billion in combined capital expenditures this year and $1 trillion in 2027 as they race to build out data centers needed to power the next generation of AI models. Hanke’s point is that those capital and operating requirements are not background noise. They are the mechanism that determines how far “the 'AI revolution' goes,” because “the cost of scarce resources that are gobbled up by AI will decide” its pace.

For executives, that matters because it reframes where risk lives. When AI spending goes vertical into power, cooling, and chip supply, the limiting factor can become real-world capacity and economics, not product marketing. If you are a board evaluating an AI strategy, you are not just asking whether a model is good. You are asking whether your company, and your region, can reliably secure enough water and electricity, and whether your unit economics can survive the infrastructure curve.

Hanke also took a harder line on how the AI narrative got sold. He described many AI visionaries as “charlatans and hucksters” and argued they are “comparing apples and oranges” when they liken AI to software. His distinction: providing AI to customers costs money and resources at run time, while software, once developed, can be sold “countless times” at “virtually no additional cost.” In other words, he is drawing a line between software distribution economics and AI inference economics. That difference can alter pricing power, margins, and how quickly AI products can scale without scaling costs.

His argument sits inside a broader skepticism he has expressed before. Business Insider notes that Hanke told the outlet in February that AI is “overhyped and potentially dangerous,” and last fall he said the AI boom could falter if Big Tech companies failed to hit their numbers. He also suggested it “might be wise to buckle your seat belt,” a sign that he sees financial fragility in the spending cycle. For corporate leaders, the subtext is clear: when capital expenditures are huge, delivery risk and demand risk both show up quickly. The question becomes not only “Will AI change work?” but “Will AI ROI match the spend cadence?”

Second-order implications ripple outward to chipmakers and investors. The source connects Hanke’s concern about cost intensity to warnings from Mark Cuban and Michael Burry. Cuban, a former “Shark Tank” investor, called it “truly scary” how central Nvidia is to the AI boom and cautioned that “all could crumble” if Nvidia gets outcompeted or makes a mistake. Burry, known from “The Big Short,” said Nvidia was “overreaching” in its campaign to strike deals that support its chip sales, and that the efforts threatened to “push the circular spending to biblical proportions.” Even if you disagree with Hanke’s job conclusion, the shared theme across these critiques is the same: the AI buildout is not just a technology story. It is a supply chain, power, and investment feedback loop.

So what should peers in strategy and finance take from Hanke’s skepticism? The job market may still change, but his argument suggests the timing and scale may be constrained by something far less glamorous than machine intelligence: the price and availability of water, power, and compute hardware. If you are an executive trying to plan headcount, re-skilling, or product roadmaps, Hanke’s frame asks a disciplined question. Before you assume AI will replace labor at scale, do the math on the resource-intensity curve that keeps the lights on for every inference request.