The benefits of the semiconductor supercycle driven by expanding artificial intelligence investment will spread from memory to makers of materials, parts and equipment, including multilayer ceramic capacitors (MLCCs) and substrates, according to a new forecast. As AI servers grow more advanced, the supply bottleneck once concentrated in high-bandwidth memory (HBM) is now spreading across the components that make up the servers, the analysis found.
Mirae Asset Global Investments presented the analysis at a webinar on the 26th under the theme "The Chip Supercycle's Second Act: Materials, Parts and Equipment." The firm said demand for MLCCs and substrates such as printed circuit boards (PCBs) would rise sharply as Nvidia's next-generation AI server platform, Vera Rubin, becomes more powerful and highly integrated than the current Blackwell.
MLCCs in particular were singled out as a key component whose usage rises quickly as AI servers consume more power. Because they supply power stably within a server, both the number of units installed and the difficulty of manufacturing them increase as AI servers become more advanced. In the global market, where Japan's Murata and Samsung Electro-Mechanics effectively hold a duopoly, inventory levels remain low despite high utilization rates. Mirae Asset said the market has entered a phase in which MLCC sales volumes and prices are rising simultaneously on the back of tight supply and demand.
A similar trend is emerging in substrates as the variety and capacity of chips installed in AI servers grow. Demand for high-performance substrates such as flip-chip ball grid arrays (FC-BGA), which connect chips to the main board, is increasing, while larger surface areas and higher layer counts make them harder to produce. When an FC-BGA grows larger, the number of substrates that can be produced from a single panel falls by about 30% to 35%, meaning that expanding actual supply in a short time is difficult even with increased capital investment.
The still-solid capacity of global Big Tech firms to invest in AI was also cited as a factor supporting the materials, parts and equipment sector. According to Mirae Asset, capital expenditure (CAPEX) by major hyperscalers this year is expected to reach $839.2 billion, an 80.6% increase from a year earlier. As of the second quarter of this year, their remaining performance obligations (RPO) stood at $2.3 trillion, about three times their annual CAPEX. With a substantial backlog of orders already accumulated for future execution, AI infrastructure investment is highly likely to continue, the analysis said.
Domestically, the firm forecast that large-scale capital investment by Samsung Electronics (005930.KS) and SK hynix (000660.KS) would produce a trickle-down effect for materials, parts and equipment companies. As investment in new semiconductor fabs gets fully underway, the range of beneficiaries could widen to include front-end equipment makers in particular. It also pointed to past periods when combined CAPEX by Samsung Electronics and SK hynix rose and the operating profit margins of domestic materials, parts and equipment firms improved in tandem.
Mirae Asset presented the TIGER AI Semiconductor Core Process ETF as a product that can ride this trend. As of the 19th of this month, Samsung Electro-Mechanics had the largest weighting at 23.96%, followed by LG Innotek (17.03%), Isu Petasys (16.75%), Wonik IPS (10.68%) and Daeduck Electronics (10.45%).
"Semiconductor-related stocks have gone through a sharp decline recently, but I think supply and demand played a big part," said Chung Eui-hyun, head of the ETF management division at Mirae Asset Global Investments. "There is a need to keep paying attention to semiconductor materials, parts and equipment companies, which are growth stocks but also have the character of value stocks with low valuations."
