Nvidia says it will build AI supercomputers with hundreds of thousands of graphics processors in the United States in the coming years.
On April 15, US President Donald Trump announced on social media that Nvidia will build AI supercomputers in the United States. The company aims to boost domestic operations as its business with China has been severely affected by the Trump administration's export restrictions.
What is a supercomputer?
Supercomputers are systems designed to perform calculations and simulations at speeds and scales far beyond the capabilities of conventional computers. They specialize in tasks that require processing many numbers at the same time, such as weather forecasting or modeling the processes taking place inside an atom.
Supercomputers use a large number of central processing units (CPUs). They are similar to the chips that run programs in personal computers or smartphones, but inside there can be tens of thousands of processors working in parallel, connected by high-speed networks.
Born in the 1960s, supercomputers are widely used in government and university laboratories in countries such as the US, China, and Japan.
What is different about AI supercomputers?
Nvidia is aiming to build large computers containing hundreds of thousands of graphics processing units (GPUs). Unlike most supercomputers, the new system focuses on training AI models, supporting chatbots like ChatGPT. GPUs are particularly useful for this task.
Nvidia's main customers are companies that provide AI services like Apple and Microsoft, as opposed to government research organizations or universities.
Compared to regular AI computers, AI supercomputers are different in level rather than type. High-performance computers equipped with GPUs are often called AI servers. They are grouped together in large computing centers called data centers.
Nvidia did not specify what makes an AI server an AI supercomputer. However, according to the WSJ, the company wanted to emphasize the high performance of the new machines, which use a large number of Blackwell chips made by the company.
Where is Nvidia's AI supercomputer made?
Nvidia has dedicated a 100,000-square-foot complex to manufacturing and testing Blackwell chips in Arizona and building AI servers in Texas. The company is working with Foxconn on a factory in Houston and with Wistron on a factory in Dallas.
The "price sensitivity" between smartphones and servers is not the same. With phones, a $100 price increase can make users reconsider, so making iPhones in the US is considered impractical. Meanwhile, large companies buying servers are willing to pay more for hardware "close to home" even though it costs more.
In addition, the iPhone production process can require hundreds of thousands of people to assemble them manually, so labor costs are very high. Meanwhile, AI server assembly is highly automated and the key factor is engineering, design, software, which is an advantage for the US.
Nvidia's choice of Texas over Silicon Valley may be due to Texas' proximity to Mexico, the current hub for AI server manufacturing. Of the servers the US imports, including both AI and non-AI servers, 70% come from Mexico, according to a report by the Taiwan Economic Affairs Bureau.
Adriana Cruz, director of economic development at the Texas governor's office, said the vast state of Texas would provide abundant energy resources and a business-friendly environment.
In January, US President Donald Trump also announced the Stargate project, which aims to spend $500 billion over four years on AI development. The first phase of the data centers is expected to be located in Abilene, Texas. Apple also plans to build a new factory in the state. Cruz said Texas "will be the hub of AI infrastructure".
However, observers say that before companies take action and achieve specific results, these "hundred billion dollar" statements need to be considered carefully.