In the manufacturing center of Foxconn located in Houston, Texas, hundreds of construction workers and engineers are working tirelessly day and night to expand the capacity for producing key components for artificial intelligence servers in the United States.
As a major supplier for Nvidia, Foxconn is undergoing a large-scale expansion with the aim of producing a significant portion of its graphics processing unit (GPU) modules and motherboards in the United States. Currently, over 90% of the critical parts are produced in Taiwan.
A senior executive at Foxconn told Nikkei Asia, “The expansion project at the Houston manufacturing center is our top priority because the majority of GPU modules and motherboards are currently produced in Taiwan.”
In a geopolitical environment that is highly complex and uncertain, manufacturing these components in the United States will help Foxconn and its customers enhance the resilience of their production capacity and supply chain, thus increasing their level of risk resistance.
This move would also be a victory for President Trump, who has been actively promoting “Made in America,” mainly through imposing high tariffs. He has mentioned imposing a 25% tariff on non-American-made iPhones.
Apple has had a minimal response to this call so far, but the response from suppliers in Nvidia’s AI server supply chain has been much quicker.
The core components of artificial intelligence servers are the GPU modules and motherboards. Currently, Foxconn and the Taiwan-based company Wistron are the sole suppliers of key components for Nvidia servers, and the research and development of these crucial components require significant engineering expertise and resources. Producing these key parts also involves close collaboration with another Taiwanese company, Taiwan Semiconductor Manufacturing Co., Ltd. (TSMC), which is also Nvidia’s largest chip manufacturing partner.
Insiders have revealed that Foxconn alone has over a thousand engineers involved in the development of Nvidia’s AI servers, handling all the work from GPU chip and computer board tray to final system assembly.
Both Foxconn and Wistron plan to expand their AI server production capacity in Texas within 15 months or even shorter, while TSMC plans to begin producing AI chips at its factory in Arizona this year, including Nvidia’s latest Blackwell chips. SK Hynix, a Korean memory chip manufacturer, is an important supplier for Nvidia, and the company is transferring the production of high-bandwidth memory (HBM) chips to Indiana. Parts suppliers such as Delta Electronics Tech, LITE-ON Technology, Qisda, and Zhen Ding, as well as assemblers like Quanta Computer, Wiwynn, and Inventec, are also establishing or expanding factories in the United States. This series of expansions indicates that building a U.S. supply chain for artificial intelligence servers (one of the most complex machines in the world) could likely become Trump’s first successful effort in relocating high-tech production to the United States.
Nikkei Asia reported that Chairman of Wistron, Lin Hsien-Ming, stated that while servers are temporarily immune to recent tariff threats, the uncertainty regarding the Trump administration’s tariff policies makes local production of artificial intelligence servers “essential.” He added that Wistron plans to complete the construction of its new factory in Texas within the next 12 to 15 months.
New Nvidia server suppliers, Shuling Technology, and United Joint Technology Inc., have indicated that they will soon finalize plans to build AI server factories, possibly deciding this month or next on whether to establish a factory in Texas.
With Nvidia announcing plans to invest $50 billion in building supercomputers in the United States and gaining support from multiple suppliers, the development of the U.S. artificial intelligence server ecosystem has begun to accelerate. This news came after Trump announced global retaliatory tariffs in early April. Nikkei Asia’s research shows that since then, at least eight Nvidia suppliers have announced new investment plans in the United States, with six planning to establish production capacity in Texas.
Nvidia’s GB200NVL72 server is currently the world’s most powerful artificial intelligence server system and one of the most complex tech products globally. It comprises over 1.2 million chips, components, and other parts, 130 trillion transistors, and copper cables up to 2 miles long. Each system weighs 1800 kilograms. Considering the volume, weight, cost of $3 million, shipping costs, and risks involved, building data centers in the United States is more practical than producing electronic products like iPhones. The United States is also where major clients such as Microsoft, Meta, Google, and AWS are rapidly establishing artificial intelligence data centers.
In comparison, producing the iconic smart phones by Apple requires more labor and many components priced at less than a dollar each. This makes manufacturing “Made in America” iPhones less feasible, despite Trump publicly urging Apple CEO Cook to move the phone production line to the United States.
TSMC is investing $16.5 billion to expand its factory in Arizona. Chairman and CEO of TSMC, C.C. Wei, stated that the company’s top priority is now to find ways to rapidly expand production capacity to meet Nvidia’s increasing demand for AI chips.
Bernstein Research estimates that by 2032, TSMC’s new investment in Arizona will enable U.S. customers to satisfy up to 50% of their chip demand domestically, compared to nearly zero in 2024.
However, this does not mean that all components will be moved to the United States. Nvidia CEO Jensen Huang told reporters in May that while it is crucial from a supply chain security and stability perspective to relocate some chip production and supply chains to the U.S., moving all server component production to the U.S. is impractical.
Nevertheless, with the most critical components and processes shifting to the U.S., it undeniably marks a significant victory for the current Trump administration.