How trends actually spread — and why predictions miss

Hype cycles, S-curves and the difference between a trend that diffuses and one that just gets talked about. A model for reading the next boom.

The question

Every year a handful of things "trend." Some become the defining industry of the decade; most vanish. Is there a way to tell a real diffusion curve from a loud but empty bandwagon before the answer is obvious?

What's happening

Trends are usually reported as sentiment — everyone's excited, so it must be real. But attention and adoption are different quantities, and they move at different speeds.

How it actually works

Most durable technologies follow an S-curve: slow start, fast middle, plateau. Hype follows a spike and often precedes the plateau by years. The gap between the two is where predictions fail.

Phase What's true What gets reported
Trigger a capability becomes possible breathless "this changes everything"
Peak of hype expectations outrun reality the loudest coverage
Trough the hard problems surface "the trend is dead"
Slope of adoption the boring work compounds less coverage, more real usage
Plateau it's just how things work now no coverage at all

Three mechanisms decide whether a trend survives its own hype:

  1. Does it have a cost curve it can ride? Technologies that improve along a predictable curve (learning rates, manufacturing volume) keep paying off. Ones that don't, stall at the hype peak.
  2. Is the demand real or borrowed? Adoption driven by genuine unmet need compounds; adoption driven by scarcity, FOMO or subsidies collapses when those fade.
  3. Can the bottleneck be moved? Most trends don't die from lack of interest; they die on a physical or economic constraint — the same supply-chain question that decides the AI boom.

So the useful question is not "is this a trend?" but "which curve is it on, and what is the binding constraint?"

A trend is a claim about the future. An S-curve is a claim about a mechanism. Trust the second one more.

What's unknown

  • How to measure adoption separately from attention, cheaply. Attention is easy to count; adoption is buried in fragmented data.
  • Whether the AI boom is on the steep part of a diffusion S-curve or still near the hype peak. Honestly: I don't know, and anyone who claims certainty is guessing.
  • Open: a repeatable scorecard for telling real demand from borrowed demand at the peak. That is the tool I actually want, and I haven't found a good version of it yet.

Reading

  • Diffusion-of-innovations literature (the original S-curve work).
  • Analyses of past hype cycles, read after the fact, to check what predicted survival.