Nvidia wants to build and sell up to half a bill of dollars of AI supercalculator equipment of American manufacturing over the next four years, with the help of Taiwan Semiconductor Manufacturing CO, AKA TSMC and its partners.
The commitment means that Nvidia will start the production of its latest generation of accelerators and Blackwell systems in the United States in the next 12 months, it also assumes that there will be enough buyers, servants or other, of this half-mini dollars of equipment made in America in the coming years.
The promise of society comes as an American business to struggle against With the Trump regime’s prices, prices outside the failure of foreign imports.
“AI global infrastructure engines are built in the United States for the first time,” CEO Jensen Huang said in a preserver statement Monday. “The addition of American manufacturing helps us to better meet the incredible and growing demand for ia fleas and supervisor, strengthens our supply chain and stimulates our resilience.”
To support this objective, Nvidia ordered over a million square feet of manufacturing space to assemble fleas in Arizona and build servers in Houston and Dallas, Texas. According to the GPU giant, these sites will possibly support the deployment of dozens of data centers across Gigawatt – or “AI factories” because Huang prefers to call them – in the United States.
The production of NVIDIA Blackwell Chips-which is based on chip-on-wafer-on-subbstrate (COWOS) packaging technology of TSMC-would have started earlier this year at TSMC Phoenix, Arizona, Chip Fabs.
NVIDIA previously used TSMC Cowos Tech Cowos to incorporate high-end bandwidth modules, but only outside the United States. At the end of last year, TSMC announced its intention to bring both its cowo and its integrated fan-out packaging technology in America as part of a collaboration with the chip packaging based on Arizona and the AMKOR test supplier, which will work with the Siliconware Industries of Taiwan Siliconware Industries (SPIL) to assemble and test its accelerators Blackwell US-Made.
To produce its increasingly hungry servers and its rack systems in America, Nvidia works with manufacturing outfits of Foxconn and Wistron contracts to run for manufacturing facilities in Houston and Dallas, respectively. Wistron appears Have a budget of $ 50 million to buy goods or land in the United States.
These factories should use the robotic platforms omavertical Twin and Isaac Gr00t from Nvidia to automate manufacturing.
The completion of the final assembly in the United States benefits Nvidia and its American customers, because the AI systems are sometimes unique and a defect or a configuration error on an imported system would be difficult to approach. The final checks carried out in the United States could save a lot of back and forth.
“We actually had chassis standards. We had food standards. We had keep-out areas (Nvidia server design spect) HGX was one of these standards, but these did not really comply with the power or even at the height of the rack,” said Patrick Moorhead Insights and Strategy, in chief strategy, Patrick Moorhead El reg. “Now, almost all these elements are tailor -made.”
At the point of Moorhead, the design of the Nvidia system has become considerably more complex and tighter tolerances in recent years while the gpu giant has passed from the semi-standard air for air for Air-Standard Air rack machines Containing more than a hundred processors cooled by liquid and GPU connected by kilometers of copper wiring.
$ 500 billion, exactly?
One wonders if Nvidia really promises to build and sell $ 500 billion in its own DGX brand chips and servers in the United States, or if this figure will include sales of Nvidia-based machines entirely made of Dell, Hpe, Supermicro and Lenovo. It is not very clear to the variable language of the advertisement of the gpu giant.
Moorhead argued that if the $ 500 billion is only sales of DGX servers, it’s quite ambitious, and if he understands Hpe and alIt is intentionally conservative to avoid being interpreted as long -term advice.
“Five hundred billion dollars out of four years of Nvidia servers, which would be very, very aggressive,” he said, adding that this would also signal a change in business model because today the ecosystem is offered a mixture of Nvidia brand servers, equipment for hypersalts and traditional HPE and Dell boxes.
Nvidia refused to comment.
GPU Titan’s commitment to us The manufacture occurs a week after reported The Trump regime planned to suspend the controls that prevented the export of the NV H20 H20 to China, after an appearance of Huang at a dollars’ dinner at the Mar-A-Lago station of the American president.
Nvidia is not the only one to move to make its products in the United States while the White House pushes companies to “reshape” their factories. Last month, TSMC revealed plans to spend more than $ 100 billion to build three new manufacturing factories in the United States, two advanced manufacturing facilities and an R&D center.
And in February, Apple announcement It plans to spend $ 500 billion in the United States, including a commitment to buy a large quantity of TSMC Arizona Fab tokens and build a server manufacturing plant in Houston – the same city where Foxconn sets up facilities with NVIDIA. It looks like a factory built by Foxconn could end up serving the two giants of technology. ®
Bootnote
Apple and Nvidia are not the only ones to look at the American capacity of TSMC. AMD late Monday revealed He had managed to validate his 5th generation Datacenter Ephyc chips, Turin code codeFor production at Fab 21 of TSMC in Arizona. AMD also said that its Epyc of the 6th generation of Venice would be built using 2 NM TSMC process technology when they arrived in 2026.