Executive Briefing: 33% of the world’s helium supply just went offline. Your AI infrastructure plan doesn’t account for it.

Source: Nate’s Newsletter Date: 2026-03-29 URL: https://natesnewsletter.substack.com/p/executive-briefing-33-of-the-worlds

Summary

A missile strike on Qatar’s helium facility took out one-third of global supply. Liquid helium containers degrade within 48 days; semiconductor fabs in South Korea and Taiwan — the primary AI chip suppliers — require helium for lithography, wafer cooling, and leak detection with no substitutes. $600-700B in annual hyperscaler capex assumes uninterrupted chip delivery. Goldman projects $1.15T in hyperscaler capex through 2027. The crisis intersects with an existing severe memory shortage.

Implications

Capital deployment / AI economic thread. This is the physical infrastructure risk that AI roadmaps systematically under-price. The concentration of helium supply in a single facility (and chip supply in a single geography) is the kind of tail risk that only surfaces when it materializes. Chinese semiconductor manufacturing gains a structural cost advantage through alternative pipeline gas sources — a geopolitical shift that reshapes the fab landscape independent of export controls.

• Pressures: any hyperscaler or AI lab with 2026-2027 compute roadmaps built on South Korean/Taiwanese fab availability needs a scenario plan; the 48-day degradation timeline means the supply shock is already in progress.
• Watch: semiconductor spot pricing and fab lead times as the helium shortage propagates through the supply chain over the next 60-90 days.