Rare Earths and the Weaponization of Supply Chains: What Hirschman Knew About Dependency

Deloitte's September update to its 2026 Aerospace and Defense Industry Outlook makes an unglamorous but important point: as the US defense sector pivots to an explicitly "AI-first" force posture, the binding constraint on that pivot is not model capability but "trusted deployment" — and deployment runs into two very physical walls. The first is materials: critical minerals and rare earths used in advanced aircraft, munitions, naval systems, satellites and secure communications remain concentrated in foreign-controlled supply chains, a vulnerability Deloitte treats as strategic rather than logistical.

The second wall is manufacturing capacity: forging and casting capability for mission-critical, long-lead components is constrained, and contractors are turning to additive manufacturing — AI-optimized digital production — to augment supply that cannot otherwise scale on the timelines an "AI-first" posture assumes. That response is economically sensible: it substitutes flexible, distributed digital capacity for a narrow, physically fixed bottleneck. It does nothing, however, for the first wall — no amount of digital production creates a rare-earth deposit or a refining line that does not exist.

That first wall is where an older economic literature is more useful than anything written about AI. In 1945, Albert Hirschman published "National Power and the Structure of Foreign Trade," arguing that trade dependency is never symmetric: a country reliant on a single supplier for an essential input cedes influence far out of proportion to that input's share of total trade value. The nominal dollar value of the rare earths in an aircraft or a guided munition is small relative to the platform's total cost — Hirschman's point is precisely that this is irrelevant. What matters is substitutability, and in the short run these inputs have almost none.

Hirschman wrote about bilateral trade between nation-states in the 1930s. Today's supply chains are networked, not bilateral, and the 2019 concept of "weaponized interdependence," from political scientists Henry Farrell and Abraham Newman, is the more precise tool. Their argument: in a globally networked production system, disproportionate power accrues not to whoever controls the most volume, but to whoever sits at the narrowest hub node the network cannot route around. For rare earths, that hub is not extraction — mining is comparatively dispersed geographically — but processing and refining, concentrated in very few facilities worldwide. Control the refining step and the mining geography becomes close to irrelevant.

This is where the AI story gets counterintuitive. AI is routinely described as a dematerializing technology — software, weights, inference cycles. But the AI-enabled defense stack Deloitte describes (autonomous platforms, dense sensor arrays, satellite constellations, secure comms) is more mineral-intensive per unit of military capability than the systems it replaces, not less. Pushing "AI-first" further does not relax the Hirschman-Farrell-Newman dependency; it tightens it, because the physical substrate scales with the AI ambition rather than shrinking as it might for, say, enterprise software.

The asymmetry this creates is not primarily a pricing problem — it would show up as a quantity constraint before it shows up as a cost one. A supplier holding the refining chokepoint does not need to raise prices to exercise leverage; withholding or throttling volume at a moment of geopolitical friction achieves the same effect with more precision, which is exactly the mechanism Farrell and Newman describe as the point of "weaponized" interdependence rather than ordinary market power.

For investors and planners in the sector, the practical implication is a reordering of where deployment risk actually sits. Additive manufacturing and other digital-production hedges address the forging and casting wall — a genuine, tractable engineering response. The rare-earth wall requires a different toolkit entirely: allied-sourcing agreements, new refining capacity, strategic stockpiles — industrial-policy instruments on multi-year timelines that no AI system, however capable, can compress. Reading Deloitte's "trusted deployment" framing without this distinction risks treating a Hirschman-style dependency problem as if it were a technology problem, which is the one framing least likely to solve it.