Wistron Corporation’s new Fort Worth factory opened July 21 and immediately began mass-producing NVIDIA’s GB300 Grace Blackwell Ultra Superchip — marking the first time the most powerful AI data center system in the world has been assembled and tested on American soil. The $700 million D1 facility, a 324,000-square-foot (30,100 square meters) greenfield plant in Fort Worth’s AllianceTexas industrial park, is Wistron’s first U.S. manufacturing facility and represents the first domestic source for the GPU systems that hyperscalers use to run — not just train — modern AI models at commercial scale.
NVIDIA founder and CEO Jensen Huang appeared at the opening ceremony alongside Wistron Chairman Simon Lin, Fort Worth Economic Development Director Jessica Rogers, and Taiwan’s Representative to the United States Alexander Tah-ray Yui. Huang personally signed the first GB300 Superchip produced at the facility — a system he described as containing roughly 1.5 million individual parts, weighing approximately 2 tons (4,000 lbs), and carrying a price tag of around $4 million per unit, according to NVIDIA’s official account of the ceremony.
“We’re producing them like phones, right here in Wistron, cranking them out in volume because the world needs all of these machines to drive the intelligence infrastructure,” Huang said at the ceremony.
What the GB300 Actually Does — and Why Inference Matters
The GB300 Grace Blackwell Ultra is not simply a faster GPU. It is a rack-scale system that integrates 72 Blackwell Ultra GPUs with 36 ARM-based Grace CPUs into a single coherent compute domain, all connected by NVIDIA’s fifth-generation NVLink interconnect at 130 terabytes per second of total bandwidth — more than enough to allow all 72 GPUs to function as a unified processing fabric rather than communicating over slower PCIe or InfiniBand connections, as detailed on NVIDIA’s GB300 NVL72 specifications page…