The Reshoring Decision

What this pack runs.

Something you sell here is made somewhere else, by your own plant or by a supplier, and the case for bringing it home arrives as two unit costs on a slide: what the supplier charges and what the unit would cost to make here. The second is higher and the move loses. That comparison is wrong twice. The supplier's price is one figure in one ledger, and freight, duty, inventory in transit, expediting, and quality sit in four others that no one adds up, because procurement is measured on the price; counted in full, the unit frequently saves money made at home. And the unit is not what decides the move. The plant is. In the nine sourced decisions none of the three reversals turned on the unit: one plant never ramped, one line made a product that did not sell, and one cost window closed. TSMC's first Arizona fab went from $12bn to $20bn and held, announced with US government support and making chips for US customers by 2025; Stanley Black & Decker's Fort Worth plant never ramped, and its tooling had no other use. This pack turns that into arithmetic. The full landed cost of a unit, less what it costs to make at home from a line that has run, times the volume sales will sign, plus what interruptions have cost and what being made here earns, is what the move saves a year. Against it: the plant, less what others pay toward it, plus the ramp and ending the current supply. The whole decision reduces to the break-even build cost, the most the plant can cost before supplying it from abroad is cheaper, and its distance from the capital request, how far over its estimate the plant can run, read beside how far the last plant the company built actually did. It stops on two halts: no one has added up what getting it here adds, and no line has made the unit.

The casebooks these cases come from.

The pack reads nine sourced decisions through the framework. They are drawn from these casebooks, each one a set of verified records on a single decision type.

A verified record has passed a mechanical claim-versus-source check, a second model's read, and human approval. The count is what exists today, not a target.