The Chosun Ilbo reported this week that Samsung Electro-Mechanics’ Package Solutions division posted first-half revenue of 1.4966 trillion won, up 41% year over year, while operating profit rose 237% to 160.2 billion won. It also reported LG Innotek package-business revenue of 935.6 billion won and operating profit of 99.6 billion won, with substrate lines running at 89% and 94% utilization respectively.
Those figures fit a wider record. Samsung Electro-Mechanics’ own second-quarter results said demand for high-value FC-BGA substrates for data centers would remain strong, while LG Innotek has publicly said it is preparing to enter the server FC-BGA market in 2026 after beginning PC-grade FC-BGA shipments to major North American technology customers. Separate reporting by MoneyToday says Samsung’s advanced FC-BGA output is already largely allocated to named customers through long-term arrangements.
The important change is that AI infrastructure has made substrate availability part of the system-level supply chain. A completed accelerator die, high-bandwidth memory stack, or server CPU cannot ship as a finished component without the package substrate designed to carry signals, distribute power, manage heat, and connect the package to the board beneath it.
FC-BGA Is Becoming a Server Build Constraint
FC-BGA—flip-chip ball grid array—is the high-density substrate used when a chip needs far more interconnects, tighter signal tolerances, and more demanding thermal and electrical performance than a conventional package can provide. The silicon die is attached face-down through solder bumps, eliminating the longer wire connections used in older packaging approaches.
For desktop PCs, package substrates have long been an important but mostly invisible part of the bill of materials. For AI systems, the scale changes. Large accelerators and server processors need physically bigger substrates with more layers and tighter manufacturing tolerances. Those attributes make defects, warpage, and yield losses more expensive, especially as each substrate supports a package containing several high-value chips and memory components.
Samsung Electro-Mechanics previously described server FC-BGA as more than four times the area of a typical PC-oriented FC-BGA substrate and as having more than 20 layers. MoneyToday likewise reported that larger AI-chip packages are increasing the difficulty of maintaining flatness, reliability, and acceptable yields. The manufacturing challenge explains why adding nominal factory floor space does not translate immediately into high-end output: production equipment must be installed, processes tuned, and each product qualified by its customer.
That is the practical reason this is more consequential than a routine components-capacity story. The constraint is not merely the number of boards a factory can fabricate. It is the volume of customer-qualified, high-yield substrates available for specific server and accelerator designs.
Samsung’s 25 Trillion Won Plan Is Broader Than FC-BGA Alone
The most eye-catching number in the Chosun report—Samsung Electro-Mechanics’ planned 25 trillion won investment through 2040—needs a qualification. Seoul Economic Daily reported in July that the company plans to invest 23 trillion won across Busan and Sejong and about $1.2 billion in Vietnam, but that plan covers both FC-BGA substrates and multilayer ceramic capacitors, or MLCCs, for AI servers and semiconductors.
In other words, it would be misleading to treat all 25 trillion won as dedicated FC-BGA capacity. Samsung is building around several AI-server component bottlenecks, including power-management parts as well as substrate packaging. Its latest earnings release highlighted long-term MLCC supply agreements with more than 10 global customers alongside plans to expand substrate supply for AI accelerators and server CPUs.
The timing also matters more than the headline total. Seoul Economic Daily reported that much of Samsung’s investment could be concentrated in the first three years, with 3 trillion won projected in 2026, 6 trillion won in 2027, and 3 trillion won in 2028. But MoneyToday reported that the meaningful effect of newly added high-performance FC-BGA capacity is not expected until 2028 because equipment buildout and customer qualification take time.
That leaves a multi-year period in which allocation and product mix may matter more than raw expansion pledges. If a supplier can choose between a low-complexity product and a customized AI package with higher margins, the AI order has a structural advantage. For system vendors, that creates pressure to secure supply contracts earlier in a product cycle and to avoid assuming that a confirmed chip allocation guarantees a complete, shippable server.
LG Innotek’s Strategy Separates Development From Scale Production
LG Innotek is pursuing a different manufacturing structure: keep advanced development and higher-value production in Gumi, South Korea, while expanding broader package-substrate production in Vietnam. The company announced in June that it would expand a semiconductor-substrate plant in Hai Phong, with construction beginning in July 2026 and completion targeted for May 2027.
LG said the Vietnam facility will make several substrate categories—RF-SiP, FC-CSP, and FC-BGA—not FC-BGA alone. Its stated plan is to use Gumi as a “mother factory” for new technology, new models, and higher-value products, while Vietnam handles more generalized semiconductor-substrate production.
That distinction is significant for anyone expecting near-term relief in leading-edge AI packaging. More total capacity can support revenue growth and ease pressure in adjacent substrate categories, but the hard part of server FC-BGA remains the high-layer-count, large-area, customer-specific product. LG’s own material says it plans to enter the server FC-BGA market in 2026, while a glass-substrate pilot line in Gumi targets commercialization in 2027 or 2028.
LG’s earlier 600 billion won Gumi agreement also deserves the same careful reading as Samsung’s investment plan. Announced in March 2025, it funds FC-BGA mass-production-line expansion but also new equipment for high-value camera modules. The company has signaled additional domestic substrate investment, yet it has not publicly broken out how much of its planned spending will be devoted to AI-server FC-BGA versus other package products.
What This Means for Enterprise Buyers and PC Hardware
For enterprise IT teams buying complete AI systems, the substrate shortage is mostly an upstream risk rather than a part to source directly. It can surface as longer delivery schedules for accelerators, server motherboards, CPU platforms, network switches, and custom AI hardware—even where the most visible chip supplier has announced ample silicon supply.
The procurement implication is straightforward: treat component allocation as a system problem. An AI-server deployment depends on advanced packaging, memory, power components, board-level assembly, cooling, and networking arriving in compatible quantities. A delivery commitment that mentions GPUs but says nothing about complete-system lead times is incomplete.
For PC enthusiasts, there is no evidence in the available reporting that this FC-BGA squeeze is about to create a broad shortage of mainstream desktop CPUs, laptop chips, or consumer GPUs. Samsung and LG are targeting higher-value server, AI, automotive, and networking products, where substrate size, layer counts, and customization are most demanding. The spillover risk is indirect: suppliers may favor premium, capacity-hungry products over lower-margin PC-oriented packages when fabrication lines are full.
The near-term winners are likely to be the suppliers that can qualify high-end substrates before new AI platforms reach volume production. The near-term constraint for server makers is more mundane: AI capacity cannot be delivered on the strength of GPU announcements alone. It also requires a steady stream of the package substrates underneath them, and Samsung and LG are signaling that those substrates are already being sold before much of the planned new capacity exists.
