AI Demand Pushes PCB Lead Times Beyond 20 Weeks; Memory Shortage May Persist Until 2028
N.R. Finch
AI capacity expansion is pushing supply-chain stress beyond chips into basic materials — PCB lead times now exceed 20 weeks and memory shortages may persist until 2028, driving infrastructure costs higher across the board.
Why are PCBs suddenly taking so long?
The bottleneck sits upstream: copper-clad laminate (CCL — the core raw material for making PCBs) is in tight supply.
Fiberglass-cloth makers are shifting to higher-end grades, but new lines have low yields, slowing the entire chain.
This means → CCL makers and PCB manufacturers are fighting over the same materials. Suppliers in Japan, Taiwan, and mainland China are raising prices nearly every quarter.
CCL makers have also begun withholding stock to reserve capacity for even higher future prices. In plain terms = upstream isn't just struggling to produce — it's deliberately throttling the pace.
What chain reaction does the 800G switch trigger?
In H2 2026 the optical-module market accelerates toward 800G, shifting PCB demand from mid-to-low-end HDI to high-end mSAP technology — a PCB manufacturing process that achieves finer circuit traces.
Very few companies can mass-produce mSAP. Key Taiwanese players: Unimicron, Zhen Ding Tech, and Chin Poon.
U.S. data-center clients are also demanding China-free supply chains, tightening mSAP capacity further. This means → mSAP lead times have surged from 6 weeks to nearly 24 weeks, with almost zero supply-side flexibility.
Why isn't the memory shortage a short-term blip?
GoldKey Technology chairwoman Tseng Hsiu-chin says memory entered a structural super-cycle starting September 2025.
The core driver: an AI server consumes roughly 8× the memory of a conventional server. HBM production — high-bandwidth memory built specifically for AI accelerators — is absorbing a huge share of fab capacity, squeezing DDR4 and DDR5 supply.
In plain terms = nobody is hoarding for speculation — fabs genuinely cannot produce enough, because their lines are being redirected to serve AI.
How steep are the price hikes and supply gaps?
GoldKey forecasts DRAM contract prices up ~30% in Q3 and NAND Flash up 60%–70%.
Customer inquiries are massive, yet actual fulfillment may be only 30% or lower; some products simply cannot be delivered.
This means → even buyers willing to pay more may not secure supply. This reflects a gap that is no longer about "waiting in line" — capacity itself cannot keep up.
What pressure is consumer electronics absorbing?
GoldKey's DDR5 share rose from roughly 50% in 2025 to 60% in 2026 and is expected to keep climbing.
PC, laptop, and smartphone makers face margin erosion from memory price hikes, while consumer demand softens as prices rise.
In plain terms = AI has claimed the memory capacity, and consumer electronics must absorb higher costs with less supply.
Can this supply-chain stress land softly?
PCB and memory supply chains are tightening in parallel; AI infrastructure expansion costs keep transmitting downstream.
Two critical checkpoints: ① whether the memory shortage eases before 2028; ② when new mSAP capacity comes online effectively.
This means → if these two bottlenecks do not open in time, everyone from cloud providers to end consumers will face sustained pricing pressure.
Content is for reference only, not financial advice.