Nvidia Is Prepaying Amkor $1.5B to Build US Chip Packaging. Packaging Is the Bottleneck.

Nvidia and Amkor struck a $1.5 billion partnership, with Nvidia prepaying to expand Amkor's advanced-packaging capacity in Arizona. Packaging, not raw chips, is the constraint on how many AI accelerators ship.

Nvidia Is Prepaying Amkor $1.5B to Build US Chip Packaging. Packaging Is the Bottleneck.

Amkor announced a multi-year, $1.5 billion strategic partnership with Nvidia to expand advanced semiconductor packaging and test capacity, with Nvidia providing a prepayment to fund a buildout in Arizona, per Yahoo Finance. The two will jointly develop high-density interconnect and heterogeneous-integration technology, the methods for combining several chips into one package. Amkor shares jumped 17% after hours.

The detail that matters is the prepayment for packaging capacity. As we covered when TSMC reported, CoWoS-class advanced packaging is the gating constraint on AI accelerator supply, ahead of wafer fabrication. A Blackwell or Rubin GPU is inert until its compute and memory dies are packaged together, and that packaging capacity has been sold out. Nvidia prepaying Amkor to add US capacity is Nvidia buying down its single biggest supply risk, and doing it outside Taiwan. The Arizona location sits alongside TSMC's own Arizona fabs and stands up a domestic packaging step that barely existed two years ago.

This is the move Nvidia has made prepaying labs and clouds, now applied to the supply chain. When you are demand-constrained by one process step controlled by a handful of vendors, you use your balance sheet to expand it and to diversify where it happens. For anyone modeling AI hardware supply, packaging capacity is now being financed directly by the chip buyer, which should ease the bottleneck heading into 2027, and US-based packaging is becoming a genuine second source to the Taiwan concentration.

Packaging is the real ceiling on AI chip supply, and Nvidia just paid to raise it on US soil. Watch Amkor's Arizona capacity timeline, because it is one of the figures that sets how many accelerators ship in 2027.