SosumonkeyOctober 8, 2026

Missed Out on Memory This Year? An Investment Area That Could Become the Next Battleground After AI Semiconductors

What to look for
The 30-minute discussion covers MLCCs' roles as dams and water purifiers, the 2 ten-thousand~3 ten-thousand units installed in AI servers, figures for order backlogs, lead times, and selling prices, and the composition of an ETF investing in the Japanese and Korean supply chain.
Why it matters
Understanding how component bottlenecks lead to a suppliers' market, pricing changes, and margin changes, and how to gauge industry conditions using order backlog ratios and lead times, can help you interpret news about AI infrastructure materials, components, and equipment for yourself.

With interest in AI infrastructure continuing, the speaker identifies MLCCs and substrates as components growing alongside semiconductors and says they will discuss investment ideas with a representative of Kiwoom Asset Management's ETF Solutions Team. They explain that AI servers contain semiconductors together with components and substrates, so increased data center investment also increases component demand. With supply unable to keep up with surging demand, bottlenecks are spreading to MLCCs and substrates, creating a suppliers' market and a trend toward improved profit margins. They explain that MLCCs are multilayer ceramic capacitors: like dams, they store and release electricity to stabilize voltage; like water purifiers, they filter noise to improve signal performance. The key technology is increasing capacitance within sizes ranging from a fingernail to a grain of sand. They explain that as GPU evolution increases power consumption, the number of MLCCs installed per GPU surges. AI servers contain 2 ten-thousand~3 ten-thousand units, more than 10 times the number in ordinary servers. Higher specifications eat into general-purpose production capacity, and components costing 4,320 dollars, only 0.06% of a rack's price, become a bottleneck, leaving substantial room for price increases. They describe substrates such as FCBGA as intermediaries connecting chips and mainboards with different terminal spacing, transmitting signals at ultra-high speed without loss. Larger areas, more layers, and advances toward glass substrates raise entry barriers and average selling prices. They report that MLCC order backlog ratios of 1.47 for Murata, 1.31 for Samsung Electro-Mechanics, and 1.72 for Taiyo Yuden exceed the 2018 peak, while lead times have extended to 16 weeks for standard products and 23 weeks for high-capacitance products. High-specification products consume 4~7 times the production capacity of ordinary products, and added capacity will become available in 2027~2029. They say that a slowdown in big tech capital expenditure coinciding with the arrival of added capacity is a risk, but one shared across all sectors. They introduce an ETF listing on October (month 10) 13, concentrated in 10 stocks and investing in the Japanese and Korean supply chain, with 25% each in Murata and Samsung Electro-Mechanics and 15% each in Taiyo Yuden and LG Innotek, among other holdings. All directional outlooks are the speaker's conditional judgments, not definitive short-term price predictions.

Timeline

8 segments

Was this helpful?

Educational information only; not investment advice.