By Yang Jie

President Trump enthused on social media this week that chip maker Nvidia would build "A.I. SUPERCOMPUTERS" in the U.S. The company hopes to boost its operations in the U.S. while its business with China has just taken a hit from Trump administration export curbs.

What machines is Trump talking about, and why will they be American-made? Here is a guide:

What is a supercomputer?

"Supercomputer" has generally referred to computers designed to perform calculations and simulations at a speed and scale far beyond what everyday computers can do. They specialize in tasks that require crunching a lot of numbers at once, such as forecasting the weather or modeling what happens inside an atom.

Traditionally supercomputers use a large number of central processing units. Those are similar to the type of chip that runs the show in a personal computer or smartphone, but inside a supercomputer there could be tens of thousands of processors working together in parallel, connected by a high-speed network.

Haven't supercomputers been around for a long time?

Yes, they go back to the 1960s. Supercomputers have been used especially at government laboratories and universities in countries such as the U.S., China and Japan.

So is the AI supercomputer Trump mentioned another one of these supercomputers?

Not exactly. What Nvidia has in mind is a big computer packed with hundreds or thousands of graphics processing units, or GPUs. Unlike most existing supercomputers, these computers focus on training artificial-intelligence models -- the kind used in chatbots such as OpenAI's ChatGPT. GPUs are especially useful for these tasks.

Nvidia's main customers are companies such as Apple and Microsoft that offer AI-powered services, as opposed to university or government research groups.

What's the difference between Nvidia's 'AI supercomputer' and other machines that power AI?

It is a difference of degree, not of kind, according to people in the industry. In recent years, Nvidia's high-end GPUs have played a key role in advancing AI, and the high-performance computers built with them are often called AI servers. These are packed together in big computer farms called data centers.

Nvidia didn't offer a precise definition of when an AI server turns into an AI supercomputer, but it likely wanted to emphasize the high performance of new machines that use a lot of its latest Blackwell chips.

Where will these machines be made?

Nvidia said it has commissioned more than a million square feet of manufacturing space to build and test its Blackwell chips in Arizona and build AI servers in Texas. The company is working with Foxconn, the world's largest contract electronics maker, on a plant in Houston and with another Taiwanese manufacturer, Wistron, on a plant in Dallas.

I've read it's impractical to build high-tech products like the iPhone in the U.S. Why are AI servers different?

Price sensitivity is a big difference. Even a $100 price increase may induce consumers to shun a smartphone. But the big U.S. companies that buy servers are willing to pay more for hardware close to home. They can absorb the higher costs in the U.S.

Also, making an iPhone may require hundreds of thousands of manual workers so the level of hourly wages matters. The assembly of an AI server is highly automated, and the critical elements are engineering, design and software. The U.S. has a comparative advantage in these areas.

Why Texas and not Silicon Valley?

For one thing, it is closer to Mexico, the current hub of AI server manufacturing. Of the servers imported by the U.S. -- including AI and non-AI servers -- about 70% come from Mexico, according to a report by Taiwan's Ministry of Economic Affairs.

Spacious Texas offers plentiful energy and a business-friendly environment, said Adriana Cruz, executive director of economic development at the governor's office. In February, representatives of Taiwan's major AI server makers met Gov. Greg Abbott to discuss expansion in the state.

The Stargate Project, another plan touted by Trump, has said it would spend $500 billion building out AI. The initial phase of data centers is planned for Abilene, Texas. Apple has said it plans a new factory in the state. Texas "will be the center of AI infrastructure," said Cruz.

Still, industry observers said the headline-grabbing announcements should be viewed with caution until the companies deliver concrete results.

This explanatory article may be periodically updated.

Write to Yang Jie at jie.yang@wsj.com

(END) Dow Jones Newswires

April 17, 2025 05:35 ET (09:35 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.