TPCA exhibition observations: AI material choices remain undecided, while equipment constrains ABF substrate expansion

Why have materials for next-generation AI servers not yet been selected?

Pure polytetrafluoroethylene (PTFE) copper-clad laminates offer the best signal integrity and are therefore favored by some customers for use in midplane and switch PCBs; however, they are more difficult to process into high-layer-count boards.
PTFE multilayer boards also require dedicated high-temperature presses operating at close to 400°C; conventional vacuum presses cannot meet the requirement. Taiwanese copper-clad laminate makers are therefore also exploring hydrocarbon resins, hydrocarbon resin and PTFE blends, and BCB, among other options.
Morgan Stanley expects materials for next-generation AI server racks to be determined by mid-2027 at the latest. If PTFE copper-clad laminates are ultimately not adopted, the bank believes the outcome would be broadly positive for EMC and TUC and slightly negative for Shengyi Technology.

Several copper-clad laminate suppliers still regard thin E-glass fabric as the main material bottleneck and generally expect shortages and price increases to continue at least through the first half of 2027; one expects them to last until the end of 2027.
Supply is tight partly because of a shortage of Japanese looms and partly because some glass fabric makers prioritize internal demand or shift production lines to LDK and T-glass fabrics that meet AI demand. One supplier said it was unwilling to use other brands of looms because of their higher defect rates and maintenance costs.

Morgan Stanley's research shows that EPM has more than 95% of the global market for ABF vacuum lamination equipment, with lead times extending to at least two years. Capacity at EPM's new greenfield plant will not come online until 2029; until then, it can only optimize existing capacity or rent idle third-party capacity.
A newly built ABF substrate plant may need at least 10 sets of EPM's advanced equipment. On that basis, the bank expects new greenfield ABF substrate capacity to begin ramping up only from 2029, with the supply shortfall worsening in 2027–2028.
Larger substrate dimensions also increase supply pressure: if they reach 230 mm × 230 mm by 2028, only 4 substrates can be cut from each panel, and a single defect on the panel could cause a yield loss of 25%.

EMC said its copper-clad laminate capacity for substrates will more than double from current levels by the end of 2028, and that it will supply several major ABF substrate customers. Morgan Stanley believes this can provide substrate makers such as Unimicron with another source of supply.
EMC's S572T copper-clad laminate for substrates entered mass production in the fourth quarter of 2026, with shipments expected to rise quarter on quarter in each quarter of 2027. Its gross margin is higher than that of PCB copper-clad laminates, so a greater share of sales could improve EMC's profitability.
Glass-core substrates face microcrack and packaging reliability problems: existing optical inspection equipment struggles to identify microcracks, which become impossible to detect after multilayer ABF lamination. Morgan Stanley expects commercial viability to remain low over the next two to three years.