AI circuit board supply tightens: loom lead times stretch to five years, constraining high-end substrate expansion

Why is capacity expansion failing to keep pace with demand for high-end substrates?

After visiting a Taipei trade show in October, Nomura found that supplies of key materials and equipment were tighter than a year earlier. Larger substrates with more layers mean that the production capacity area consumed per unit is growing faster than shipments.
The divergence in supply and demand is central to the assessment: Prismark believes the overall ABF substrate market is oversupplied, but high-end substrates are in short supply; Nomura expects material and equipment constraints, combined with higher specifications, could intensify the high-end shortage over the next few years.
Nittobo expects its capacity for thick T-glass fiberglass cloth to more than double in FY28 from FY26, but the company believes demand is too large for it to meet alone, and its market share could still decline by FY30.

Looms may be the most severe constraint on fiberglass cloth capacity expansion. Nomura observed lead times of about five years for Toyota Industries looms, up from about three years a year earlier; fiberglass manufacturers interviewed believe Tsudakoma looms have yet to meet industry standards.
Other equipment also has long waits: lead times for laser drilling equipment have risen from about one year to three years, substrate exposure equipment from about one year to three to five years, and ABF vacuum lamination presses from less than six months to two and a half years.
The loom shortage also affects fiberglass cloth used in PCBs. Trade show respondents generally expect shortages of low-end E-glass and monthly price increases to continue at least through the first half of 2027; Nomura believes that if manufacturers shift capacity from low-dielectric-constant fiberglass cloth to E-glass, price pressure on the latter could ease, but supply of the former would tighten.

Copper-clad laminate manufacturers told Nomura that copper foil supply appears to present no problem in 2026, but could tighten over the next few quarters: the ramp-up of new AI platforms is driving demand for high-end copper foil, while manufacturers' conversion of old production lines to high-end products is reducing low-end supply.
For example, Co-tech stopped producing low-end HTE and RTF copper foil in September 2026 and plans to convert its production lines; the company also plans to add annual capacity of 10 thousand tonnes of HVLP3 and HVLP4+ in 2027–2029.
Nomura believes high-end copper-clad laminate manufacturers will also gain pricing power. Its base-case assumption for EMC is a further price increase of 10%–20% beginning in the fourth quarter of 2026. Low-end material prices also depend on the reallocation of fiberglass cloth capacity among different products.

Nomura expects new platforms from the second half of 2027 through 2028 to make greater use of M9K1/K2 copper-clad laminates paired with M4/M6, as GPU and ASIC performance improves and signal-loss requirements become more stringent; the current generation primarily uses M8K1/K2 paired with M4/M6. Board structures will also shift further toward high-density interconnect (HDI) or hybrid HDI and multilayer boards.
Optical modules are another source of demand: near-package optics (NPO) boards require at least 16–18 layers of mSAP HDI, compared with 12–16 layers for 1.6T pluggable modules and 10–12 layers for 800G modules. Nomura believes growth in optical module boards could improve margins at Zhen Ding Technology and Compeq.
Nomura reiterated its buy rating on substrate maker Unimicron, citing supply bottlenecks that support pricing and technology upgrades such as embedded components and multilayer core boards that could widen leading manufacturers' competitive advantages.

Glass-core substrates are attracting more attention because of their low thermal expansion, low loss and suitability for large-format manufacturing, but Nomura believes commercialization is still some way off: mismatched thermal expansion between glass and resin causes microcracks and reliability problems, while solutions such as buffer layers add process complexity.
Resin approaches for high-speed copper-clad laminates are also diverging. PTFE can lower the dielectric constant, but TUC says it cures at about 380°C for more than six hours; HCR cures below 250°C in less than three hours, but has a higher dielectric constant.
Blending PTFE with HCR is a compromise, but processes including lamination and drilling, as well as dimensional stability, remain challenging. Some board manufacturers visited by Nomura believe the difference in electrical performance between HCR and blended PTFE is not particularly pronounced.