Binding Constraint
The input that actually limits output, which is usually not the one being discussed and often not the one anyone is spending on.
Reviewed July 16, 2026Stable
When not to use it
- Where several constraints bind simultaneously, in which case identifying one is misleading and the system needs joint relaxation.
- As a forecast. A constraint identified today is local to current conditions and moves.
- Where the apparent constraint is a policy choice rather than a physical limit, which is a different problem with a different remedy.
Reach for something else instead
- Constraint succession modelling — plan against the sequence of bottlenecks rather than the present one.
- Shadow price estimation — where the system is formalisable, compute how much relaxing each constraint is actually worth.
- -
Read more on the blog
- 1 to 2% of the chips, about 30% of the tokensExport controls were designed to constrain compute and have. The measure that moved instead was usage, and the two have separated sharply.
- The good numbers all came from obligationsTerritory 8 closes. Across ten subjects the reliability of a figure tracked whether someone was legally required to produce it, and nothing else.
Further reading
- International Energy Agency (2025), Energy and AI — the infrastructure lead times against which compute demand is set.
- International Federation of Robotics, World Robotics 2025 — an installed base showing where physical automation was and was not constrained.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Extrapolating the current bottleneck, which is how forecasts fail in a predictable direction.
- Treating an eased constraint as still binding because the coverage has not updated.
- Assuming capital relieves a constraint that requires physical manufacturing time.
At a glance
FieldFoundations
Originlinear programming, shadow prices
Testwould twice as much change the output
Signalprices that look irrational relative to apparent importance
Traprelieving one constraint promotes the next
DifficultyBeginner → Intermediate
Where this sits
A starting point. Nothing needs to come before it.
0Levelstarting point
0Needs firstnothing
0Opens upnothing further
1Areastays here
Computed from the prerequisite graph, not assigned. How this works