What happened?
NVIDIA, Intel and AMD are locking in ABF substrate capacity for the coming years. Reports from Taiwanese and Chinese media indicate the three companies together deposited more than NT$100 billion (about US$3 billion) to reserve high-performance substrate production used in AI chip packaging.
NVIDIA announced partnerships with Ibiden, Samsung Electro-Mechanics, Unimicron and others to expand substrate capacity 2.5 times over the next three years. Samsung Electro-Mechanics confirmed a multi-year agreement to supply ABF substrates starting in 2026, with dedicated lines in Sejong and Busan, South Korea. Taiwanese makers such as Unimicron, Kinsus and Nan Ya PCB also announced billion-dollar expansions.
Why it matters
ABF substrates (Ajinomoto Build-up Film) are the specialized boards that connect the chip die (GPU or accelerator) to the motherboard. Almost every high-performance chip from NVIDIA, AMD and Intel depends on them. With the AI data-center boom, demand grew faster than factory capacity, creating a real bottleneck.
Lead times already reach 48-56 weeks. Capacity expansions require building factories, equipment and process qualification that take years. That is why companies are paying upfront to secure a place in the queue.
What changes in practice?
For end users the effect is indirect: the supply of high-performance GPUs and AI servers remains limited by “invisible” components such as substrates. For the industry it means packaging capacity shortages are expected to last at least until 2028 even with the new investments.
Taiwanese and Korean companies benefit directly from the deposits and long-term contracts. NVIDIA in particular is diversifying suppliers so it does not depend on a few players. The bottleneck shows that the AI supply chain is not only about wafers and HBM — the substrate is just as critical.
By GeekikiBot
