When a friend asked Tyler Cowen what could stop or significantly slow the advance of AI, his answer was short: 'Not much.'
Cowen, Holbert L. Harris Professor of Economics at George Mason University and Faculty Director of the Mercatus Center, made the case in a recent essay that the AI revolution has already cleared its most significant hurdles — and that the pace of progress is driven less by capital markets than by the internal logic of the technology itself.
The bubble fear is real. The conclusion drawn from it is wrong.
Cowen does not dismiss the possibility that AI is, at least in part, a bubble. He points to a concrete data point: an AI-oriented hedge fund called Situational Awareness lost $35 billion over the space of two weeks in July. He calls it a scare — not a verdict.
His argument is historical. In the early 2000s, the dot-com bubble burst. In the mid-2000s, the housing bubble burst. In both cases, few recognized the bubble until the bust arrived. And yet Amazon, eBay, and Google kept growing. Homes kept being built. The financial wreckage did not erase the underlying productive reality.
'Bankruptcy, it turns out, doesn't stop progress,' Cowen writes.
The analogy he reaches for is railroads — a sector that produced spectacular financial collapses 150 years ago and also produced the infrastructure that industrialized a continent. The investors who lost money were not wrong about the technology. They were wrong about the timing and the returns. The tracks remained.
What the record actually shows.
Cowen acknowledges the financial exposure plainly: most major AI companies are spending billions while counting on future revenue that has yet to materialize. He declines to offer investment advice or assign probabilities to bankruptcy scenarios. His point is narrower and more durable — that 'the foundation upon which AI models are being built will continue to exist, and to grow,' regardless of which firms survive.
That is not a prediction about stock prices. It is a claim about the persistence of productive infrastructure once it exists.
The Signal's read.
Cowen's framework is a useful corrective to two symmetrical errors: the booster who treats every valuation as justified, and the skeptic who treats every correction as a refutation of the technology itself. Neither position follows the evidence.
What the historical record does support is this — markets are imperfect allocators of capital in the early stages of transformative technology, but they are not the only mechanism by which that technology advances. The knowledge, the models, the trained engineers, the physical compute infrastructure: these do not disappear in a Chapter 11 filing. They get acquired, repurposed, or rebuilt.
For readers who care about free enterprise and innovation, that is the more important signal. The question worth asking is not whether the next AI unicorn survives its Series D. It is whether the regulatory and legal environment allows the underlying technology to keep compounding. On that question, Cowen is conspicuously optimistic — and the burden of proof, given the historical pattern he cites, rests with those who believe this time is categorically different.
