Chokepoint
A chokepoint is an input that a few holders control and everyone downstream needs. When it tightens, the whole system slows, and the holders decide who gets served. Allocation, not price, clears demand.
exmxc gives every chokepoint one of two statuses:
- Binding: the input is limiting output now.
- Tight: the input is scarce but not yet the limit.
Chokepoints sit in all four of the Four Forces of AI Power. In Compute, they include EUV lithography, leading-edge chipmaking, advanced packaging and high-bandwidth memory. In Energy, they include grid connections, transformers and switchgear, gas turbines and firm nuclear power. In Alignment, the clearest example is government permission to sell a model. In Interface, it is control of the default assistant.
Control of a chokepoint is what the Scarcity Multiplier in sPEG is built to price. But a chokepoint is a state, not a permanent property: holders can choose to widen it. When the owner of a chokepoint adds capacity, the scarcity that justified a premium starts to unwind. The Scarcity-Durability Framework exists to tell the two cases apart.
The current, sourced map of AI chokepoints, with each holder, status and dependency, is maintained at AI Lattice.
Related: Four Forces of AI Power, Signal Brief: Four Tells in One Tape Week and Firm Capacity.