According to reports, in the face of the Trump administration's policy expectation that it may impose high import tariffs on high-end AI hardware, Nvidia is working with manufacturing partners to establish a complete AI server supply chain in Texas, USA. This strategic move aims to circumvent potential trade barriers and is also trying to build end-to-end local manufacturing capabilities, from chip production to server assembly.
Nvidia has joined forces with industry giants such as TSMC, Foxconn, Wistron, Amkor and SPIL to advance this plan. Previously, the Trump administration planned to impose high import tariffs on expensive AI hardware, and in order to avoid this risk, NVIDIA chose to work with partners to build an AI server supply chain in the United States. From chip production and packaging to server assembly and deployment, the entire supply chain will be localized.
Foxconn and Wistron plan to build two server assembly plants in Houston and Dallas. Currently, construction work is about to start, and actual assembly is expected to begin in the next 12 to 15 months. This is NVIDIA's first attempt to complete the full process assembly of AI infrastructure in the United States, aiming to strengthen the resilience of the manufacturing industry and meet the growing global demand for AI. Nvidia expects the factories to produce $500 billion worth of AI devices over the next four years.
To improve the design and operational efficiency of the new plant, NVIDIA will use its cutting-edge technology. With the Omniverse platform, NVIDIA can simulate factory operations and optimize production processes in advance. Isaac GR00T is used to develop an automated robot system that integrates AI into every aspect of production to achieve intelligent and efficient production.
In terms of chip manufacturing, TSMC has started production of Blackwell processors at its No. 21 factory near Phoenix, Arizona. However, at this stage, the chip still needs to be shipped to Taiwan, China for packaging. But that soon will change as Amkor and Silicon plan to build more than 1 million square feet (about 92,900 square meters) of state-of-the-art packaging and test facilities in Arizona. At that time, the whole process of chips from production to packaging will be further localized.
Pictured: Nvidia will build a $500 billion AI server empire in the United States (Source: Tom's Hardware)
Nvidia CEO Jensen Huang said, "For the first time, the core of the world's AI infrastructure is being built in the United States. Manufacturing in the U.S. will help us better meet the huge and growing demand for AI chips and supercomputers, strengthen our supply chains, and improve our ability to respond to risks.”
For NVIDIA, building AI servers in the U.S. is significant. This move localizes the AI supply chain, and although labor costs are high in the United States, the impact on costs is not significant given the high added value of AI servers. And, as the Dallas and Houston plants ramp up the use of AI and robotics, the actual cost of assembling the system is expected to be further reduced through continuous optimization.
Of course, there are some challenges to this massive project. For example, the issue of water use in factories is of great concern. If the open cooling tower is used for water spray cooling, the water consumption should not be underestimated; The rational solution is to use heat exchangers and closed-loop systems to reduce water consumption. In terms of power supply, with the development and application of small-scale nuclear power generation technology, the power problem is expected to be alleviated in the next few years.
The promotion of NVIDIA's plan will reshape the supply chain ecology of the AI industry, drive the development of related industries in the United States, and attract more resources to invest in the AI field. At the same time, it will also intensify competition in the global AI market, prompting other companies to accelerate the pace of innovation and push the entire AI industry to a higher level.
