chokepoints.ai
SUBSCRIBE
10 layers580 nodes2,376 dependencies9 chokepoints112 bottlenecks6,500+ companiesnode size = companies identified

METHODOLOGY

How we classify the AI compute stack

580 markets. Four dimensions, each scored 0–5 from cited evidence. One deterministic rule. 9 chokepoints.

The thresholds are published below. The rule is code — no judgment call sits between a node's scores and its label.

The four dimensions

Every market is scored 0–5 on each dimension. A score with no citable source is floored at 1 — never the benefit of the doubt.

DimensionThe questionScores 5Scores 3Scores 1
C · ConcentrationWho controls supply?One firm holds ≥80% of the flow (HHI ≥ 5,000)Top three firms hold 50–80%Fragmented — top three under 25%
M · MoatHow long until a qualified alternative supplier exists?Decade-class know-how or ecosystem lock-in; qualification takes 24+ monthsReal capital or IP barrier; 6–24 monthsCommodity process; switching takes under 3 months
D · Downstream coneIf supply halved for a year, what breaks?The whole stack strainsOne major layer is impairedNiche — local impact only
R · Route-aroundWhat does substitution cost?No route inside 5 years at any capex2–3 years plus heavy requalificationDrop-in alternatives exist today

The rule: chokepoint, bottleneck, commoditised

Labels fall out of the scores mechanically. First rule that matches, wins.

CHOKEPOINT

Rule: C ≥ 4 and M ≥ 3 and D ≥ 3 and R ≥ 4. Meaning: concentrated, defended, load-bearing, and non-routable — all four at once. Whoever holds one can durably tax the stack.

BOTTLENECK

Rule: D ≥ 3 and R ≥ 3, but fails the chokepoint bar. Meaning: the stack strains here, but control is too diffuse or too weakly defended for anyone to durably tax the flow.

COMMODITISED
337 of 580

Rule: C ≤ 2 and M ≤ 2. Meaning: fragmented and undefended. Price does the talking.

The remaining 122 markets sit between the bars — tracked and re-scored, not labelled.

What survives the test

580 markets scored

9Chokepoints
112Bottlenecks
337Commoditised
122Tracked, unlabelled

How a score is set

  1. 1

    Evidence first. Every dimension score carries at least one dated, cited source. No source, no score above 1.

  2. 2

    Three draws, median stands. Each dimension is scored three times independently; the median is the score. No single pass sets a label.

  3. 3

    Adversarial challenge. A separate pass attacks every score — it can hold a score or cut it, never raise it.

  4. 4

    The rule, applied. The label is computed from the thresholds above. Nothing is hand-tuned at the label step.

Every node scored on concentration, moat, downstream reach and routability, then adversarially re-verified. Rescored 15 Jul 2026 · 990 dimensions challenged.

What we publish, what stays on the desk

Published

  • the label and the thresholds behind it
  • a one-line why for every chokepoint and bottleneck
  • the top suppliers at each market

On the desk

  • the dimension scores themselves
  • supplier share estimates
  • direction of travel
  • the full tracked roster

Full scorecard, owner shares, supply edges and the full tracked roster are in the desk letter.

One company per issue, free, straight to your inbox.

GET THE BRIEFING