Samsung Electro-Mechanics has signed a multi-year agreement to supply advanced package substrates for Nvidia’s compute chipsets and for the high-speed networking switches that link them inside data centers. Deliveries begin this year.
The two companies will jointly invest in expanding dedicated capacity at the Sejong and Busan plants, the Korean sites that house the firm’s substrate business. The partnership was disclosed at Nvidia’s Berlin conference, alongside agreements with a handful of other capacity partners.
These parts are the electrical and mechanical bridge between a bare die and a circuit board. For accelerators they have to carry enormous current while holding signal integrity across thousands of connections, which is why suppliers embed multilayer ceramic capacitors and silicon capacitors directly inside the substrate.
Volume sent to Nvidia over the multi-year term is expected to grow at a double-digit annual average, and brokerage IM Securities read the arrangement as Nvidia underwriting capacity because it sees demand through 2028 and real risk of shortage.
The move also widens a relationship that had been limited. The Korean firm already supplied substrates for Groq’s third-generation language processing unit, but this is the first formal acknowledgment of broad Nvidia supply.
The strategic shift is who pays. Components that once sat low in the cost stack now draw direct investment from the buyer, which treats them like wafer starts.
