Samsung Electro-Mechanics Invests $5 Billion to Expand ABF Substrate Production, NVIDIA Secures Dedicated Capacity with Prepayment
nashnova research
Samsung Electro-Mechanics announced a combined $5 billion investment in Korea and Vietnam to expand ABF substrate production, with Nvidia prepaying to lock dedicated lines — a sign that packaging bottlenecks are forcing the AI chip supply chain to place heavy bets years ahead.
Where does the $5 billion go?
The Sejong plant in Korea gets KRW 4.27 trillion to expand ABF substrate — a high-end packaging material that connects AI chips to circuit boards — with mass production targeted for September 2028.
A Vietnamese subsidiary receives an additional KRW 2.51 trillion to scale up substrate lines, with equipment due by end of April 2028.
The combined KRW 6.78 trillion (≈$5 billion) is a single-product investment record for Samsung Electro-Mechanics.
This means → the company is elevating ABF substrates from one product among many to a company-level strategic priority.
Why is Nvidia paying upfront?
Nvidia accounts for the largest share of prepayments; the two sides finalized dedicated-line terms in H1 2026.
Samsung Electro-Mechanics confirmed in early 2026 that it would supply ABF substrates for Nvidia's NVSwitch chip, formally entering Nvidia's supply chain.
In plain terms = Nvidia is not placing orders and waiting — it is wiring money ahead of time to lock capacity under its own name. Whoever controls substrate capacity gets first access to next-generation AI chip production.
How do the two plants divide the work?
Sejong is positioned as the core base for high-value, high-difficulty products, serving top-tier customers.
The Vietnam plant handles volume expansion; this round of investment matches the scale of the original 2013 factory build, underscoring the magnitude.
This means → Korea handles the cutting edge, Vietnam handles scale — together they cover the full cycle from flagship to volume ramp for AI servers.
What are the risks?
The mass-production milestone is September 2028 — more than two years away, during which AI chip roadmaps and customer demand could shift.
The utilization rate of Nvidia's dedicated line is the critical variable: if Nvidia's product cadence slows or pivots to alternative packaging, that capacity could sit idle.
This reflects a deeper tension in the AI semiconductor supply chain: expansion must be committed two to three years in advance, but demand-side certainty doesn't stretch nearly that far.
市场有风险，内容仅供研究参考，不构成投资建议。