Samsung Electro-Mechanics to supply Nvidia with AI chip substrates

The hardware backbone of AI

While the world focuses on the software and models running on artificial intelligence, the physical hardware requires specific components to function at high speeds. One of these critical parts is the semiconductor substrate.

A substrate acts as a middle layer that provides the necessary electrical connections between a semiconductor chip and the printed circuit board. For high-performance computing systems, these connections must be incredibly reliable and capable of handling high-speed data transmission without losing signal integrity.

As AI models become more complex, the demand for these advanced substrates has surged. Samsung Electro-Mechanics has positioned itself to meet this demand by signing a multi-year agreement with Nvidia to supply these components for AI processors and high-speed networking chips.

Scaling production in South Korea

To meet Nvidia's requirements, Samsung Electro-Mechanics is moving beyond its current output. The company will initially fulfill orders from its existing facilities in South Korea before moving into a larger expansion phase.

The agreement includes joint investments to upgrade existing plants and build new facilities in the cities of Sejong and Busan. These investments are designed to ensure that the supply chain can keep up with the rapid pace of AI hardware deployment. According to the Korea Herald, deliveries are expected to start as early as this year.

By 2028, production for Nvidia is expected to represent a substantial portion of Samsung Electro-Mechanics' total annual packaging-substrate capacity. This indicates a deep integration between the two companies, where Samsung becomes a primary backbone for Nvidia's hardware ecosystem.

A multi-partner strategy

Nvidia is not relying on a single source to meet its production goals. The company is coordinating with several key partners to ensure a steady flow of components. In addition to Samsung Electro-Mechanics, Nvidia is working with Ibiden in Japan and Unimicron in Taiwan.

This multi-partner approach is intended to increase substrate production capacity for Nvidia's products by 2.5 times over the next three years. By diversifying its supply chain across different regions and companies, Nvidia reduces the risk of production bottlenecks that could delay the release of new AI hardware.

The financial specifics of the deal, including the total volume of deliveries and the monetary value of the contract, remain undisclosed. However, the growth in the sector is evident in Samsung's recent performance. Their package solutions division reported revenue of 771.6 billion won, or $583 million, in the second quarter, which was a 37% increase compared to the previous year.

Broader trends in AI hardware

The partnership between Samsung and Nvidia is part of a wider trend of hardware companies seeking to optimize how AI interacts with the physical world. While Samsung focuses on the internal connections of the chips, other companies are looking at how those chips interact with external sensors.

For example, LG Innotek is currently developing cooperation with Nvidia in the robotics space. They recently showcased a 3-megapixel camera module and a multimodal machine-vision system that uses Nvidia Holoscan Sensor Bridge technology. This tech allows sensors to send data directly to AI computing systems, which speeds up how robots process information from their surroundings.

LG Innotek plans to further develop visual perception systems for humanoid robots and robotic arms, with plans to show off next-generation solutions in 2027. These developments show that the AI hardware ecosystem is expanding from the data center into physical robotics, requiring a sophisticated web of substrates, sensors, and high-speed networking.

The road to 2028

The long-term outlook for these partnerships suggests a massive build-out of infrastructure over the next few years. Samsung's commitment to expanding its footprint in Sejong and Busan suggests that the demand for AI-ready hardware is not a short-term spike but a sustained shift in manufacturing priorities.

For the reader, this means that the