The most dangerous number in AI right now isn’t a valuation multiple – it’s the ratio of experimental-to-production deployments

Last Thursday I walked into the WJNO studio knowing I'd get pushback. That was the point. When a Drive Time producer called me the week before, they wanted the standard segment: two minutes on markets, a prediction, out. I told them I needed six minutes minimum and I'd probably make their audience angry. They gave me eight.

The segment was about our AI Implosion thesis — the idea that the current AI investment cycle is structurally similar to the dot-com bubble, not in degree but in kind. Same pattern of capital flowing faster than revenue can justify. Same gap between what's promised in pitch decks and what's delivered in production.

Before I get into what happened after, let me be clear about what I'm not saying. I'm not saying AI is a fad. I'm not saying the technology doesn't work. I'm saying the investment thesis surrounding it — the valuations, the hiring, the infrastructure spend — is disconnected from the actual rate of enterprise adoption. And that gap has consequences.

What I Said on Air

The core of my argument comes down to three data points our sentiment model has been tracking since Q3 2025. These aren't proprietary numbers — they're public if you know where to look. But nobody's connecting them.

  1. Enterprise deployment rates have plateaued. After a surge in 2024, the percentage of Fortune 500 companies moving AI projects from pilot to production has flatlined. They're still experimenting. They're not operationalizing.
  2. Developer sentiment has shifted. This is the qualitative signal our model picks up earliest. The tone on internal forums, in conference hallway conversations, in hiring patterns — it's moved from "this changes everything" to "this is useful for specific things." That's a meaningful shift.
  3. The capex-to-revenue ratio is historically unprecedented. The amount being spent on AI infrastructure versus the revenue it's generating makes the 1999 telecom buildout look conservative. That's not hyperbole — run the numbers.

Every major correction in the last 30 years was preceded by the same four words from analysts: 'This time is different.'"

— From the Pulse Index Sentiment Model, Q1 2026 Report

What Got Cut

Radio is a blunt instrument. Eight minutes sounds like a lot until you're in the chair. What I didn't get to say — and what matters most — is the timeline.

Our model doesn't predict crashes. It predicts sentiment inflection points. And the current data suggests we're 12–18 months from a significant repricing event in AI-adjacent equities. Not a crash. A repricing. The distinction matters because it affects how you position.

A crash is indiscriminate. A repricing is selective. The companies with real revenue from AI — the ones actually solving problems at scale — will be fine. The ones running on narrative and runway will not.

A repricing is not a crash. Repricing events are selective — they separate companies with real revenue from companies running on narrative. Our Q1 white paper breaks this down by sector.

The Three Questions That Kept Coming In

After the segment aired, the station's phone lines lit up for the next hour. Most callers fell into three camps.

"Are you saying I should sell my NVIDIA shares?"

No. I'm not a financial advisor, and this isn't investment advice. What I'm saying is that the sentiment signals that preceded every major tech correction of the last three decades are present right now. What you do with that information is your decision.

"Isn't AI adoption still growing?"

Yes — but growth rate and growth trajectory are different things. A plane that's still climbing but decelerating is going to level off. The question isn't whether AI is growing. It's whether it's growing at the rate the current valuations require. Our data says no.

"What about the jobs AI is replacing?"

This was the most common question by far, and it's the one I'm least equipped to answer in a radio segment. The labor displacement question is real and it's serious. But it's also separate from the investment thesis. Technology can be genuinely transformative and still be overvalued as an investment. Those two things coexist all the time.

I'll be writing more about each of these threads in the coming weeks. The next white paper — due mid-April — will cover the enterprise adoption data in detail, including the specific indicators we're watching and what the historical comparisons actually show.

If you want the full picture before it publishes, premium subscribers get early access and the underlying data tables. Otherwise, check back here. The blog is where I think out loud, and I've got a lot more to say.