Is the AI Bubble Close to Popping? Wrong Question, Says This Strategist

Every AI-related stock in the market seemed to get repriced twice within a single week recently — hammered the day the Federal Reserve held rates steady, then up more than 3% on the Nasdaq the very next session, as though nothing had changed in between. Strategist Jim Bianco’s view is that the whiplash itself is the market asking the wrong question. Whether AI is “a bubble,” in his framework, was never really about how far these stocks move in a single day.

His actual test comes from history: every major technological bubble, from nineteenth-century railroad expansion to the fiber-optic buildout of the early 2000s, ends the same way — the world builds meaningfully more of the underlying infrastructure than there turns out to be demand for. US railroad mileage peaked in 1916 and remains roughly 40% below that peak today. Telecom companies laid more fiber-optic cable by 2001 than the world would actually use for years afterward. By that same measure, Bianco argues, AI infrastructure isn’t close to that point yet: only a small share of businesses are currently using AI in ways that generate measurable, attributable revenue or cost savings, and demand for computing capacity is still running ahead of available supply rather than behind it.

The risk he actually worries about isn’t valuation multiples or sentiment swings — it’s a sudden leap in computing efficiency. If a new model allows the same chips and the same power draw to deliver several times the useful output, every data center currently under construction or recently completed becomes instantly oversupplied, without a single new watt of capacity being built. That’s roughly what briefly spooked markets when a competing model claiming exactly that kind of efficiency leap surfaced a few weeks earlier.

He also draws a distinction that’s easy to collapse into one category but shouldn’t be: generative AI, the conversational tools most people already use directly, versus agentic AI, software that completes entire tasks end-to-end without a human at the keyboard — which he considers the far less mature, and ultimately more consequential, half of the story.