Samsung affiliates steer nearly 8 trillion won of a 12.26 trillion won chip plan to Vietnam

Take Samsung Electro-Mechanics' 6.78 trillion won out of the 12.26 trillion won total and Samsung Electronics' testing plant in Thai Nguyen comes to about 5.48 trillion won ^[18]. Its $4.09 billion price converts to the same sum at the package's own exchange rate ^[18]. Adding the 2.51 trillion won of substrate capacity bound for Vietnam brings the Vietnamese side to 7.99 trillion won, about 65 percent of the package ^[19]. Korea gets Sejong's 4.27 trillion won, roughly 35 percent, all of it substrates ^[20]. Samsung Electronics puts no new won into Korea here ^[20]. What it gets at home, in the Korea Herald's account, is test capacity the move could free up at its Korean sites for advanced memory products ^[5].

Only the substrate maker writes the value split into its plan. Its flip-chip ball grid array substrates connect high-performance processors to circuit boards, and demand is growing as AI servers need larger, more complex packages ^[17]. Sejong is to take the high-value, technically demanding products and Vietnam the growing global orders ^[11]. For memory, the split is an outside reading. Joo Seung-hwan, a professor at Inha University's Graduate School of Manufacturing Innovation, read the plan as an apparent effort to hold high-bandwidth memory and other high-value products at Cheonan and Onyang and send more of the conventional memory work to Vietnam, according to the Korea Herald ^[6]. Samsung has not disclosed how testing volumes will be divided ^[7].

The financing terms are the better puzzle. At the package's exchange rate, the reinvested-profit share of the test plant is about 3.45 trillion won that Samsung's Vietnamese operations have yet to earn ^[21]. The other 39.15 trillion dong, about 2 trillion won, is the only part not tied to those earnings ^[21]. Vietnam pays for most of its own test lines, or rather, Vietnamese profits still to come do ^[8]. Global customers are helping fund Samsung Electro-Mechanics' expansion and have committed to buying from it over the medium to long term, the Korea Herald reported ^[12].

Vietnam's memory testing comes first on the calendar. The first test line is due in the fourth quarter of 2027, with a second in 2028 ^[3]. Sejong begins substrate mass production in September 2028, and the Vietnamese substrate site follows 21 months later, in June 2030 ^[10][22].

From here the Vietnamese test lines could absorb growth in conventional DRAM and NAND testing while the Korean sites keep their work and add HBM. They could instead take over some testing now done in Korea; the Korea Herald says it remains unclear which ^[15]. Or the substrate split could blur before 2030, since Samsung Electro-Mechanics already makes advanced substrates in Vietnam ^[11]. We think the value split describes the substrate plan accurately and is a reasonable but unproven account of memory testing. The counter-case is cost. The 103 Vietnamese engineers trained in memory testing and equipment operation at Cheonan and Onyang, according to industry sources, fit a transfer of existing Korean work as well as an addition to it ^[13]. The thesis is wrong if Cheonan or Onyang test lines are cut instead of refilled with HBM work, or if high-bandwidth memory turns up on the Thai Nguyen lines ^[6].

Local sourcing is a separate test. "Expanding production facilities overseas and building a local semiconductor ecosystem are two different things," an industry source said, as quoted by the Korea Herald ^[16].