At its Berlin developer conference on Wednesday, Nvidia named six manufacturers it is leaning on to build the less glamorous half of an AI machine: the boards, packages, test gear and light engines that sit around the processor.
The list is Amkor Technology, IBIDEN, King Yuan Electronics, Samsung Electro-Mechanics, Siliconware Precision Industries and Unimicron Technology. Each covers a different bottleneck.
Where the money went
The largest single figure is a $1.1B advance to King Yuan Electronics for testing capacity. New lines in the United States are due in volume next April, with Singapore following in February 2027.
Nvidia told investors it would lift substrate supply 2.5 times within three years. IBIDEN’s Ono plant in Japan is being pushed to three times its current output, with extra volume arriving in the first quarter of 2028. Samsung Electro-Mechanics joins the substrate roster immediately from existing Korean lines and will expand at Sejong and Busan.
In Korea, Amkor and Nvidia aim for more than a 300 percent increase in interposer wafer capacity by the end of 2028. In Taiwan, Siliconware will raise optical engine output fifteenfold by mid-2027 and double that again by late 2028. Unimicron adds substrate volume from the second quarter of 2027.
The bill behind the promise
These are not gifts. A regulatory filing shows Nvidia’s purchase and capacity commitments reached $279B by late July, up $160B in a single quarter, mostly for memory and factory space. The filing concedes the company may not be able to walk away if demand sags.
That is the trade Nvidia is making. It buys certainty with cash advances and multi-year reservations, and it carries the risk if the buildout slows.
