Bottom line first: the ramp in AI server volume is pushing high-end PCB demand structurally towards higher layer counts and thicker copper — the share of 20-layer-plus high-multilayer mainboards and thick-copper power boards is rising together. The real bottleneck is not whether the boards can be built but lamination cycles and interlayer registration: each step up in layer count stacks up more lamination passes, more accumulated material expansion and tighter registration tolerance, which makes front-loaded engineering review a decisive factor in lead time.
Demand structure: mainboard and power board upgrade together
The PCB content of one AI server is concentrated in three places: the CPU/GPU mainboard, the switch chip substrate and the power distribution board. The first two pursue high layer count and low loss; the third pursues thick copper and current carrying. These two routes place different demands on a fab — the former tests lamination and impedance control capability, the latter tests plating uniformity and resin fill capability. Procurement should evaluate the two categories separately at RFQ stage instead of applying one set of metrics to every board.
The distinction matters because the two capabilities are rarely equally strong in the same plant. A supplier that runs high-multilayer digital boards well may still be average at thick-copper filling, and the reverse is equally true. Splitting the enquiry makes the comparison honest and keeps the quotation comparable.
Three engineering costs of rising layer count
- More lamination cycles: boards above 20 layers usually require multiple lamination passes, and every pass introduces material expansion and stress, so warpage risk scales with layer count.
- Harder interlayer registration: different material expansion coefficients shift the inner-layer patterns, so expansion compensation factors have to be built into calculation tables by material group, rather than applying rule-of-thumb values.
- Tighter thickness tolerance: additional layers accumulate total board thickness tolerance, which affects later connector insertion and thermal compression, so upper and lower thickness limits have to be agreed at design stage.
None of these three is a surprise on its own; the difficulty is that they arrive together. A stackup that gains layers late in the programme changes the lamination sequence, the compensation table and the thickness budget at the same time, which is exactly the point at which schedules slip.
Separate evaluation points for thick-copper power boards
The key question for a thick-copper board is not how thick the copper can be made, but current-carrying temperature rise and plating uniformity. At the same 3oz copper thickness, an uneven distribution of copper in the holes will push local temperature rise above the limit. When evaluating a supplier, ask for thick-copper plating uniformity data and current-carrying test reports, and state the acceptance lower limits for hole copper and surface copper explicitly. Without those two documents the copper thickness on the drawing says very little about what the finished board will do under load.
Procurement actions: write front-loaded review into the lead-time plan
- Front-load the review: give the stackup, material group, impedance targets and thick-copper requirements in one package at RFQ stage, so that repeated redesign after prototyping is avoided.
- Dual-source layout: qualify suppliers for high-multilayer boards and for thick-copper boards separately, so that a single supplier under capacity pressure does not block both lines at the same time.
- Reserve first-article confirmation separately: the first-article confirmation cycle for large high-multilayer boards should be planned on its own and not compressed into the same week as the volume production schedule.
Shenzhen Lingxin Technology Co., Ltd. operates a high-end board line aimed at AI computing, covering 1–44 layers, finished copper up to 13oz (outer-layer base copper up to 18oz) and a maximum panel size of 610×1100 mm, with lead times as fast as 24-48 hours, depending on product complexity. For an engineering review or a quotation, send Gerber data and stackup requirements to info@ipcba.com.cn or call +86 138 2310 9230.
Send your Gerber files and stackup requirements; a 24-hour expedite channel is available.