Interview: From afterthought to bottleneck, circuit boards move to the heart of AI hardware

The AI server motherboard has crossed a threshold: it is no longer a passive carrier but a system-level design variable. When layer counts exceed 20 and board thickness approaches substrate-scale dimensions, the PCB co-determines signal integrity, thermal dissipation, and power delivery. Chip architects must now iterate with PCB engineers pre-tape-out — the same logic that pulled CoWoS packaging into front-end design in 2019 is propagating downstream along the value chain. Supply-chain concentration remains acute. High-layer ABF substrate capacity sits with Unimicron (Taiwan, China) and Ibiden (Japan), while mainland players like Shennan chase mid-tier volume. If export controls extend from photoresists to high-frequency CCL laminates and specialty resins, AI hardware BOM costs face a 15-25% structural premium. Competitively, Ibiden has locked in NVIDIA's next-gen platform substrate share; AT&S captures European sovereign-compute policy tailwinds; mainland firms remain structurally trapped — capacity without design voice. Within 18 months, expect three shifts: PCB makers entering chip roadmap reviews for the first time; low-Dk CCL materials becoming the next chokepoint; and 2-3 vertical-integration M&A deals in the substrate-PCB space, repricing the industry from per-layer to per-system-value economics.