America Lost the Packaging, Not the Chip

In the second week of October the largest contract chipmaker in the world signed a five year, two billion dollar agreement with an American foundry to make silicon interposers in upstate New York, with volume production due in the first half of 2028. An interposer is the slab of silicon that sits under several chips and wires them to each other, and it is the structural piece of the advanced packaging used in every high end processor.

Days earlier a Massachusetts research institute broke ground on a hundred and eleven thousand square foot microelectronics and packaging building, noting in passing that the United States performs around three percent of the world's advanced chip packaging.

From the Battlefield to the Balance Sheet

Three percent is the number that should have been driving the policy all along. The national conversation about semiconductors has been about wafers, because wafer fabs are expensive and visible and easy to photograph. Packaging is the step after the wafer, where individual chips are cut, stacked, connected and sealed into something a circuit board can use, and it is the step that left the country almost entirely.

It left for ordinary reasons. Packaging was historically labour intensive and low margin, so it followed cheap labour to Asia in the 1970s and 1980s and the industry stopped thinking about it. Then it stopped being low margin. Modern high performance processors are not single chips at all, they are several pieces of silicon mounted together on an interposer with high bandwidth memory beside them, and the packaging determines the performance of the finished product as much as the transistors do. The step that was offshored because it was unimportant became the step that decides whether a part exists.

For defense the consequence is specific and awkward. A chip fabricated in a trusted American fab and then sent abroad to be packaged has left the controlled chain at the point where the most physical access is possible, and the alternative has been a very short list of domestic facilities working at low volume and high cost.

What makes this particular agreement interesting is that it is commercial. It is one foundry buying capacity from another, priced and scheduled as a business arrangement rather than funded as a subsidy. That is the first evidence that domestic packaging can clear on economics rather than on appropriations.

The Dual-Use Reality Check

This is the most completely dual-use step in the entire chain, and the defense share of it is negligible. The demand pulling these investments is artificial intelligence accelerators, where packaging capacity has been the binding constraint on output for two years, and where every additional interposer converts directly into a sellable part.

Defense benefits as a passenger. A domestic packaging base built for commercial processors can run defense lots at the margin, and the qualification work that follows is the same kind of narrow, expensive, slow step this newsletter keeps finding at the bottom of every supply chain. The reverse is not true: a packaging capability built only for defense volumes would never reach the scale that makes it cheap.

The piece still missing is memory. High bandwidth memory stacks are assembled in Asia, and no amount of domestic interposer capacity changes where the memory comes from. A supply chain that is domestic through the interposer and foreign at the memory is not yet a domestic supply chain.

The Capital Signal

The capital signal is that advanced packaging has become a traded commodity with real contracts behind it, and that the American share of it is going up from a very low base on commercial terms. Expect more agreements of this shape and expect them between companies rather than between a company and a government.

The structural read is to look at packaging capacity rather than wafer capacity when assessing whether a semiconductor plan is credible. A fab announcement without a packaging arrangement describes half a product. Watch the ramp dates, because every one of these facilities is scheduled for 2028 or later and the demand is now, which means the shortage persists for two more years regardless of what is announced. Watch too for whoever qualifies first on defense specific lots, because in packaging as in everything else in this sector, the qualified line is the asset and the building is just where it sits.

Signal: The step America actually lost was packaging, not fabrication, and it is coming back on commercial contracts because artificial intelligence demand made it worth money.