Samsung Electro-Mechanics will invest ₩6.8 trillion (roughly $4.7 billion) in substrates for AI semiconductors, splitting the buildout between its Sejong complex in South Korea and its manufacturing site in Vietnam.

The figure, reported by finance.biggo.com, marks one of the largest single substrate commitments by a Korean component supplier to date. It lands squarely on the industry's most capacity-constrained layer of the AI supply chain: the advanced packages and interposers that carry accelerator silicon from Nvidia, AMD and hyperscaler in-house designs.

Demand for AI substrates — including flip-chip ball grid array (FC-BGA) packages and the glass and silicon interposers used in high-bandwidth-memory-stacked accelerators — has outstripped supply since generative AI workloads pushed accelerator shipments up sharply in 2023 and 2024. Substrate makers including Ibiden, Shinko Electric and TSMC's subsidiary lines have all announced capacity programs; Samsung Electro-Mechanics' ₩6.8 trillion pledge puts it in the same weight class as the Japanese leaders for the first time.

The dual-site structure matters commercially. Sejong, south of Seoul, gives the company engineering proximity to Samsung Electronics' packaging operations and to Korean customer qualification cycles. Vietnam — where Samsung Electro-Mechanics already runs plants in the Yen Phong industrial zone near Hanoi — offers lower operating costs and a hedge against trade friction, since substrate output there serves export customers outside Korea's tariff perimeter. Running both sites simultaneously shortens the time between groundbreaking and qualified volume supply, which is the binding constraint for AI hardware makers right now.

Samsung Electro-Mechanics has historically derived most of its revenue from multilayer ceramic capacitors and camera modules for smartphones. The AI substrate program shifts its center of gravity toward datacenter components, where average selling prices per unit run far higher than in mobile and where customer concentration among a handful of accelerator vendors creates both revenue scale and qualification risk.

The company has not yet disclosed, per the report, a year-by-year capex schedule, a wafer- or panel-level capacity target, or a revenue figure tied to the investment. Those details will determine whether the ₩6.8 trillion converts into leading-edge substrate supply by 2026–2027, when Nvidia's Rubin generation and successor accelerator platforms are expected to require still larger package bodies and finer line-and-space substrates than current Rubine-era parts.

Competitively, the move tightens the race in a segment where Ibiden and Shinko have held technical leadership on build-up substrates for high-end accelerators, and where TSMC's CoWoS packaging bottleneck has forced chip designers to queue for interposer capacity since 2023. A qualified Korean supplier at scale gives accelerator vendors a second sourcing option they currently lack, and it pressures Taiwanese and Japanese incumbents on pricing as new capacity comes online.

Watch for Samsung Electro-Mechanics to follow with a phased timetable and customer qualification milestones; the pace at which Sejong and Vietnamese lines reach volume output will decide whether the company captures AI substrate share or simply adds capacity into a market that rivals are also expanding.