Data 2026-03
Scenarios
Five stress cases over the AI-compute supply chain, each dated by when it was last reviewed and by the age of its newest input.
Five stress cases over the AI-compute supply chain, each dated by when it was last reviewed and by the age of its newest input.
No review recorded. This is the oldest substrate in the instrument and the analysis has not been re-read against current cable status. The figures on this page are human-authored and sourced individually; they are not produced by the cascade engine. Newest input dated from Newest dataVintage across the module's cable and incident records, '2024-Q4' — src/lib/data/scs-cable-data.ts. The pre-consolidation page badged this as '2024', which understated the module's own newest value.. Review state for all five analyses is on the index.
Submarine cable corridors carry the cross-region traffic that AI training runs spanning Asia and Europe depend on, that US labs need to deliver inference to Asian users, that Gulf data centres use to federate with European replicas. Where those corridors narrow to single chokepoints, cable disruption becomes a continuity question for AI compute.
Submarine cables carry roughly 99% of intercontinental data traffic; 3 chokepoint zones (Bab-el-Mandeb, Malacca, Luzon Strait) concentrate that traffic onto a small number of routes. The Feb–Mar 2024 Red Sea event removed an estimated ~70% of Asia–Europe corridor capacity for weeks while repair was delayed indefinitely in contested waters.
The engine models firm-to-firm supply propagation, not cable topology — a severed submarine cable is not an edge in the supply-chain graph. What it can run is the downstream face of a cable disruption: the hyperscalers whose cross-region service the severed corridors carry (Amazon, Microsoft, Google, Oracle) losing that connectivity, propagated through their supply chain. The cable-routing, chokepoint-redundancy, and rerouting-latency analysis on this page is the part the engine does not reach.