Five stress cases over the AI-compute supply chain.
Regional disruption scenarios
Explore regional disruptions to chip production, electricity supply and subsea cables, with maps of exposed infrastructure and analysis of supply-chain dependencies.
Plate I. Two dates per analysis, on one axis. The band is the newest input to the analysis, drawn at the precision its source gave: a quarter-dated figure is three months wide, a year-dated figure twelve, because a narrower mark would claim precision the source did not supply. The vertical tick is the last recorded human review, drawn as a point because that date is exact. The bar between the tick and its due date is the review cadence; amber is the part that has run past. A hatched track means no review has ever been recorded for that analysis: there is no tick and no window, and the trigger evaluation has no baseline to measure change against. On those tracks the short line with end bars beneath the hatch runs from the newest input to today, which is how long the analysis has stood unreviewed. The vertical rule is today.
Review triggers before the next scheduled pass
11 watches across 5 analyses; 5 of them are manual.
The weekly scan compares each analysis’s watch scope with facility records written since its last review. Machine-checkable watches report fired or clear. Change classes absent from the platform’s tables, including strikes, cable cuts, and published market-share revisions, report “not evaluable”: a predicate that cannot fire cannot support “clear.” An analysis without a recorded review lacks the comparison baseline, so every watch remains unevaluable until a review is recorded.
New or edited Taiwan facility records: capacity, operator, or status change at a leading-edge site. (watched: TW)
New or edited facility records at the substitution foundries the capacity-gap analysis prices; a line that comes up or slips moves the gap. (watched: US, KR, JP, CN)
A tier-1/tier-2 publication of company-level leading-edge foundry share, which would supersede the 2021-vintage CSET country aggregate the concentration input currently uses. (manual: no table on this platform records this class)
New or edited facility records across the Gulf, Israel and Turkey: the exposure set the range rings are drawn against. (watched: AE, SA, BH, QA, KW, OM, IL, TR)
A regional strike, interception episode or air-defence deployment change that requires the incident history or defence context to be reviewed. (manual: no table on this platform records this class)
New or edited Saudi facility records: a new data centre changes which grid zone the exposure analysis has to cover. (watched: SA)
A published change to Saudi transmission topology or substation assignment. Which data centre shares a named substation with which petroleum facility is the strongest evidence class in the table, and a reassignment moves a pair into or out of it. (manual: no table on this platform records this class)
New or edited Japanese or Korean facility records: memory, wafer, materials or equipment sites. (watched: JP, KR)
A revision to published DRAM or HBM market share. How much of world memory supply sits in the Korean fabs is stated as a share, so a revision changes the size of the claim this analysis makes about them. (manual: no table on this platform records this class)
New or edited facility records in the cable-landing and hub regions the route analysis depends on. (watched: SG, HK, AE, EG, IN)
A cable cut, completed repair or new system entering service may change the routes available in a case. Review the affected-cable set and historical account when those conditions change. (manual: no table on this platform records this class)
Continue
These analyses pin named stress cases to the supply-chain graph. For arbitrary node removal and propagation outside those cases, use the cascade engine.
