Serenity turns Digitimes lead-time data into a supply-chain map, tagging the listed companies behind each choke point in the AI hardware stack. The longest wait is ABF substrates at roughly one year, the resin-film base layer that carries advanced chips, made by names like Ajinomoto (which supplies the film), Ibiden, Unimicron and Nan Ya PCB. HDI boards and multilayer boards run over six months (Zhen Ding, Victory Giant, Compeq, Meiko). CCL, the copper-clad laminate underneath, is over six months with "severe" material shortages, tightening suppliers from Elite Material and Shengyi to material makers like Resonac and Mitsui Kinzoku (copper foil) and Nittobo (glass cloth). Passive components, MLCCs, chip resistors, tantalum and aluminum capacitors, span 15-20 weeks to over a year, spreading demand across Murata, TDK, Taiyo Yuden, Yageo, Vishay and others. Serenity flags nuance: many of these parts mix commodity uses with AI-datacenter grades, so not every shortage is a pure AI play. Still, the mapping makes a point the chip headlines miss, the tightest constraints on building AI servers now sit in the unglamorous layers: substrates, boards and passives.
Why it matters · The tightest AI-server constraints have moved downstream to substrates, boards and passives, the unglamorous, year-long-lead-time names may be the cleanest read on real buildout capacity.
Worth asking · Do substrate and passive-component makers offer a cleaner AI supply-chain bet than the crowded chip names, or is the space too fragmented and commodity-mixed to play?