🤖 AI Summary
The first Nvidia Blackwell GPU wafer has been produced at TSMC’s new Fab 21 in Phoenix, Arizona—the first time a cutting‑edge data‑center GPU wafer has been manufactured on U.S. soil. Blackwell, Nvidia’s fastest architecture for AI and HPC, was previously only made in Taiwan; the Phoenix facility is running the same 5nm process used for those chips and is designed to eventually churn out up to 20,000 wafers per month. The plant’s buildout began in 2020, went into production in 2024, and is part of a multi‑fab campus planned to reach 3nm, 2nm and even 1.6nm nodes over the coming years.
This milestone matters technically and geopolitically: it increases supply resiliency for hyperscale AI demand, brings advanced node capability to U.S. soil, and underscores the scale of investment (initially $12B, later supported by CHIPS Act funding and loans) required to localize semiconductor manufacturing. Practically, the wafer step is only the start—chips still require packaging, testing and system integration (e.g., into GB200 NVL72 systems)—but onshoring wafer production of Blackwell should help Nvidia and customers blunt capacity bottlenecks and reduce risk from concentrated overseas fabrication.
Loading comments...
login to comment
loading comments...
no comments yet