The chip supply chain is what actually gates this quarter's model releases, and this week it surfaced twice: once in Korea, once in a courtroom.

SK Hynix priced the largest foreign IPO in US history on July 10, raising $26.5 billion, and the roadshow message was blunt: the company is under pressure to build new fabrication plants on US soil rather than keep expanding at its Icheon and Cheongju sites in South Korea. SK Hynix makes the high bandwidth memory chips (HBM), the stacked memory that sits next to a GPU and feeds it data fast enough to keep it busy. Nvidia's B200 and GB200 systems, the hardware Anthropic, OpenAI, and Google all train on, are bottlenecked as much by HBM supply as by the processor itself. A raise this size, with US fab pressure attached as a condition, is Washington extending the same chip strategy it has applied to Taiwan Semiconductor Manufacturing Company: pull the supply chain onto US soil so an export control on the finished chip cannot be routed around by moving the memory stack somewhere else. Separately, Apple filed suit against OpenAI on July 9, alleging trade secret theft tied to hardware and interface work, the same week reporting emerged that OpenAI withheld evidence in the New York Times copyright trial. Neither dispute is about a model's benchmark score. Both are about who controls the pipes, silicon on one side, training data on the other, that decide which lab can keep shipping.