July 28, 2026
While the spotlight stays on GPUs and HBM, one set of numbers is quietly redrawing the PCB industry's map: AI server PCBs and substrates accounted for just 7% of the global market in 2024. According to Prismark's latest forecast presented at TPCA this year, that segment will reach $15.6 billion and a 16% share in 2026 — and close to 30% of the entire market by 2030. This is not a cyclical upswing; it is a structural reallocation. This article breaks down what the super cycle actually means for engineering teams, across demand, hardware, materials, and supply chain.
The fuel is unambiguous. Combined capital expenditure from the big four cloud providers (Amazon, Microsoft, Google, Meta) ran about $470 billion in 2025 and is projected to surge roughly 80% to around $860 billion in 2026. About 65% of that goes directly into AI infrastructure hardware, and bare‑board PCBs (including IC substrates) represent roughly 2.5% of hardware cost — implying an addressable AI‑specific PCB market above $15 billion in 2026 alone. Against a global PCB market that grew 15.8% in 2025 and is forecast to grow another 12.5% in 2026, nearly all of that growth traces back to this single thread.
AI rack architecture is pushing the PCB from supporting component to core system asset. The signature change is replacing intra‑rack cabling with high‑layer‑count boards: NVIDIA's Vera Rubin NVL72 compute tray adopts a 22+22 layer midplane on an M8 hybrid stackup, cutting cable assembly from about two hours to roughly five minutes. The switch tray moves to a 32‑layer board with M8+‑class material; Groq's LPU tray reaches 52 layers on M9‑class laminate; and the Rubin Ultra Kyber rack may carry a midplane built from six 28‑layer sections — 168 layers combined. PCB value per xPU climbs from a few hundred dollars in the 8‑GPU generation to nearly $2,000 in the Kyber generation. For R&D teams, 224G signalling, ultra‑low‑loss materials, and high‑aspect‑ratio lamination are shifting from flagship‑only requirements to the baseline for high‑speed product lines.
The flip side of surging demand is material scarcity. The rigid CCL market grew more than 20% in 2025, with the high‑speed category up nearly 50% — the steepest of any segment. On the supply side, price hikes keep landing: major Japanese resin and CCL makers announced increases of up to 30%, glass fabric and HVLP‑grade copper foil rose 15–20%, and Taiwanese CCL makers followed with their own adjustments. Two practical implications for design teams: first, price assumptions on high‑speed laminates go stale within six months, so shorten quotation validity and build material‑inflation headroom into cost models; second, lead times on ultra‑low‑loss materials are volatile, making second‑source and grade‑equivalent planning more important than ever.
This cycle is not lifting all boats. The research view is blunt: AI infrastructure demand is concentrating in a small group of advanced board and substrate makers, while conventional PCB segments stay largely flat and absorb rising material costs. Top‑tier fabricators posted revenue growth of 30% and up in 2025, almost entirely on the server/networking line. For teams running mainstream product lines, the pragmatic read: advanced capacity and high‑speed materials will keep being pulled toward AI orders, so prototype and production scheduling for non‑AI programmes needs earlier capacity and material lock‑in than before.
For readers in Taiwan, the regional numbers matter: Taiwan's domestic PCB output returned to growth in 2025 and is forecast to grow about 14% in 2026 — among the fastest of any region. More than 80% of AI server PCB expansion investment since 2025 is concentrated across Taiwan and China, because the complete ecosystem — fabricators, CCL, equipment, chemistry — already sits here. From high‑layer‑count lamination and backdrilling to volume experience on M8/M9‑class materials, what Taiwan's supply chain gains in this cycle is not just orders, but a voice in defining next‑generation specifications.
The AI super cycle's rewrite of the PCB industry is structural: product mix tilting toward HDI and high layer counts, materials upgrading to the ultra‑low‑loss generation, and capacity plus investment concentrating in fewer hands. For engineering teams and decision‑makers, the numbers worth tracking are not any single quarter but three structural signals — PCB value per xPU, high‑speed CCL supply and pricing, and where your own product lines sit on this new map. The next round of spec definition and order allocation will be decided where those three lines intersect.
