Wistron Opens 324,000 sq ft NVIDIA AI Factory in Fort Worth, Texas

Wistron opened its first U.S. manufacturing facility in Fort Worth, Texas on July 22, 2025. The 324,000-square-foot greenfield plant, representing a $700 million investment, currently produces NVIDIA GB300 Grace Blackwell Ultra Superchips and is preparing a second line for the NVIDIA Vera Rubin Superchip. The facility has already created over 500 jobs, with plans to reach 1,000 by the end of 2025. NVIDIA CEO Jensen Huang and Wistron Chairman Simon Lin both attended the opening ceremony.

What happened

Detail	Fact
Facility size	324,000 sq ft (greenfield)
Investment	$700 million
Jobs created so far	500+, targeting 1,000 by end of 2025
Products manufactured	NVIDIA GB300 Grace Blackwell Ultra Superchip; NVIDIA Vera Rubin Superchip (coming)
Production target	Tens of thousands of boards per month
NVIDIA U.S. manufacturing commitment	Up to $500 billion in AI platforms

Wistron, a Taiwanese contract manufacturer, cut the ribbon on its first U.S. plant (internally called D1) in Fort Worth on July 22, 2025. NVIDIA founder and CEO Jensen Huang appeared alongside Wistron Chairman Simon Lin at the ceremony, which also drew Taiwan government officials and local Fort Worth civic leaders.

The plant runs two manufacturing cells. One is already producing the NVIDIA GB300 Grace Blackwell Ultra Superchip. The second cell is being set up for the NVIDIA Vera Rubin Superchip, NVIDIA’s next-generation platform. Wistron says D1 is scaling toward tens of thousands of boards per month this year.

Huang described the GB300-based system as “the most powerful AI supercomputer in the world,” adding that each unit contains 1.5 million parts, weighs two tons, and costs $4 million. He signed the first unit produced on site at the ceremony.

How Wistron built a factory inside a simulation first

Before a single machine went on the floor, Wistron built a full digital twin of the facility using NVIDIA’s Omniverse platform, Nemotron and Cosmos frontier models, Metropolis libraries, and the PhysicsNeMo open framework. Engineers used the virtual factory to validate assembly line layouts, fine-tune production workflows, and train workers on standard operating procedures, all before the physical plant came online.

This approach, sometimes called “simulation-first manufacturing,” is becoming more common in high-complexity electronics assembly. It reduces costly real-world trial and error and shortens the time between breaking ground and full production capacity.

Why does this matter for the AI industry?

The Fort Worth facility is one concrete piece of NVIDIA’s stated commitment to manufacture up to $500 billion in AI platforms inside the United States. Until recently, most advanced AI hardware was assembled in Asia. Moving production to the U.S. shortens supply chains for American cloud providers and data center operators, reduces geopolitical risk, and creates a domestic workforce with skills in high-precision electronics manufacturing.

Huang framed it in broad terms at the ceremony: “Just as agriculture is a fundamental infrastructure of society, in the future these AI factories will be a fundamental infrastructure of society.” The comparison is deliberate. AI compute, in NVIDIA’s view, is not a product category but an infrastructure layer, like roads or power grids.

For businesses evaluating AI infrastructure, domestic production of GB300 and Vera Rubin systems could eventually mean shorter lead times when ordering high-end GPU clusters, though enterprise buyers will still be working through NVIDIA’s partner channels rather than buying direct from Wistron.

Our take

The numbers here are real and large. A $700 million factory, 500 jobs on day one, and $4 million per unit being “cranked out like phones” is not vaporware. Whether you read Huang’s infrastructure-era framing as visionary or theatrical, the physical building exists and is shipping product.

What is genuinely interesting is the digital twin approach. Simulating an entire factory before construction is a strong signal of where industrial AI is going: not just inside the product but inside the process that makes the product. Businesses in manufacturing, logistics, or complex operations should watch this closely. The same tools (Omniverse, PhysicsNeMo) are available outside of NVIDIA’s own projects.

For most of our clients, this news sits one or two steps upstream. But if your business depends on GPU availability for AI integration projects, a more robust domestic supply chain is a genuine medium-term positive. We cover more developments like this on the Lumien AI news feed.

What to do about it

1. If you are sourcing GPU compute for AI workloads, ask your cloud or hardware provider about lead times on GB300-class hardware over the next 6 to 12 months. Domestic production may change availability windows.
2. If your business involves physical production or assembly, look at NVIDIA Omniverse and PhysicsNeMo as practical simulation tools, not just research projects.
3. Watch for Vera Rubin Superchip availability announcements. The second production cell at D1 coming online will signal when that hardware enters the supply chain.

The practical takeaway: U.S.-based AI hardware production is no longer a policy talking point. It is a working factory in Fort Worth shipping $4M supercomputers at volume.

Source: NVIDIA Blog