The Moving Bottleneck

A system moves at the speed of its slowest necessary part — and the moment you fix that, the bottleneck jumps somewhere else.

The Future of Technology · Lesson 13 · 7 min read

A company spends a fortune doubling the speed of one station on its assembly line — and total output barely moves, because a different, slower station was the real limit the whole time. The big forces behave the same way: pour resources into the part that is already fast and you get almost nothing back. So how do you find the part that actually controls the outcome — and where it will go next? Hold that question; it is one of the most practical models for predicting where progress flows.

A system is only as fast as its slowest necessary part

When several things must all happen for an outcome, the slowest or scarcest one — the bottleneck — sets the pace. Improving any other part does almost nothing, because the bottleneck still caps the result.

The deep forces are complements: AI needs compute, compute needs energy, real-world deployment needs data and trust. Because each truly needs the others, the system’s speed is set by whichever complement is currently scarce — not by the ones that are already abundant.

Relieve a bottleneck and it moves

Here is the twist that trips everyone up: when you fix the slow step, the constraint does not vanish — it relocates to the next-slowest step. Progress is a game of chasing the moving bottleneck.

So “we solved X, now nothing is holding us back” is almost always wrong. Solving X just hands the baton to whatever is now scarcest.

As AI models got good and compute got cheap and abundant, the binding constraint shifted: first the model itself was the bottleneck, then it moved to energy (power for all those data centers) and data (the right data to use), and as those ease it shifts again toward integration and trust (getting it safely into real workflows and past regulation). The scarce thing today is not the scarce thing tomorrow.

Find the bottleneck to predict where value flows

This is the payoff. Effort, money, and attention rush to whatever is the current bottleneck, because relieving it is where the biggest gain is. So to forecast where the next opportunities — and the high prices, and the breakthroughs — will appear, ask: given how cheap the other forces are getting, what is now the scarce complement?

The honest caveat: because bottlenecks shift, this is a snapshot skill. You do not answer it once; you keep re-asking as each constraint is relieved and the next one steps forward.

Trace the moving bottleneck of “deploying AI usefully”:
1. Early on the model is the bottleneck (not capable enough) → effort and money pour into better models.
2. Models get good and compute gets cheap, so the bottleneck moves to energy (powering the compute) and data (the right data) → resources rush there.
3. As those ease, it shifts again to integration and trust (safe deployment, regulation) → attention follows.
Each time the scarce complement is relieved, value and effort jump to whatever is now scarce. Predicting that jump — not admiring the part that is already cheap — is the whole skill.