As HBM and memory tighten, substrates are also in short supply ... Big Tech says, "We’ll fund it, just give us the volume" - 매일경제

As the artificial intelligence (AI) semiconductor market expands rapidly, the supply chain is facing new bottlenecks. While High Bandwidth Memory (HBM) was once the hardest component to obtain, shorta... Read original →

Surging AI chip demand is shifting bottlenecks from HBM to foundational packaging substrates—especially FC-BGA. Fabricating these requires ultra-fine build-up layers and low-loss dielectrics, with technical barriers rivaling some front-end processes. As TSMC, Samsung, and Intel adopt EUV at 3nm, stringent thermal and signal integrity specs for substrates are weeding out smaller players. LG Innotek, though late, leverages ABF expertise from RF SiP to potentially enter NVIDIA and Google’s tier-2 supply chain. Crucially, Big Tech’s prepayment-for-capacity model is flipping industry power dynamics: fabless firms now proactively secure midstream resources instead of waiting passively. Over the next 18 months, substrate shortages will delay GDDR7 and SOCAMM adoption and accelerate vertical integration among suppliers in Taiwan, China, Japan, and Korea. Regions lacking domestic substrate capacity risk tangible AI deployment delays if geopolitical tensions escalate.