Morgan Stanley: MLCC Evolving from Commodity to Strategic Resource in the AI Era
Claire Weston
Morgan Stanley says AI servers consume 10–15× more MLCCs than conventional servers, while industry capacity grows just 10–15% a year — a supply-demand mismatch that could last years, with prices already surging.
Why do AI servers devour so many MLCCs?
MLCCs — multi-layer ceramic capacitors, the tiny "reservoir" components on circuit boards that keep power delivery stable — are needed in vastly greater quantities for AI. Nvidia's GB300 rack uses ~320,000 units; the next-gen Vera Rubin platform jumps ~1.8× to roughly 570,000 per rack.
This means → a single AI server consumes the MLCC equivalent of 10 to 15 conventional servers.
Morgan Stanley estimates AI-driven MLCC demand will more than quadruple between 2025 and 2030, while total industry capacity expands just 10–15% per year. In plain terms = demand is on a rocket; supply is on an escalator.
How far have prices already moved?
Channel checks show distributor prices have risen 200–300% since Q1 2026; spot prices for some specs are up 6–10×.
Major suppliers raised distributor prices 20–30% QoQ in Q2 2026. Some cancelled volume discounts and refused to honour Q1 low-price orders.
This reflects a shift from "fight for share" to "choose your customer" — having supply matters more than having a price.
How is this cycle different from 2017?
The 2017–18 MLCC super-cycle was driven by 5G smartphone upgrades and EV ramp — a cyclical demand shock. Once capacity caught up, prices fell back.
This time the bottleneck is on the supply side: AI servers require ultra-low ESR and ESL — equivalent series resistance and inductance, which determine how fast a capacitor can respond to nanosecond-scale transient currents. Consumer-grade and automotive-grade lines simply cannot produce these specs; retooling does not bridge the gap.
In plain terms = last time it was "everyone wants to buy, temporarily not enough to sell"; this time it is "even if you max out every old factory, you still can't make what AI needs." New lines take ~2 years from investment decision to mass production, equipment is highly customised, and strict qualification barriers limit who can enter.
Who supplies this market?
The high-end AI-grade MLCC market is a Japan-Korea duopoly with a combined share of ~85%: Murata at ~45%, Samsung Electro-Mechanics (SEMCO) at ~40%.
Murata's AI-server orders run at roughly twice its capacity; utilization is 90–95%. President Nakajima said demand for cutting-edge MLCC products will "significantly exceed expectations for at least the next three years."
SEMCO's Tianjin plant is running at full load. Morgan Stanley lifted its SEMCO target price from KRW 920,000 to KRW 2,560,000, maintaining Overweight. This means → the bank thinks SEMCO's earnings revisions are just getting started. Chinese maker CCTC (三环集团) is entering the domestic server supply chain, but high-end share remains in Japanese and Korean hands.
How big can the market get — and how long will the shortage last?
Morgan Stanley forecasts global MLCC shipment value rising from $14.7 bn in 2025 to $24.3 bn in 2028 — an 18% CAGR, well above the prior decade's 6.5% trend.
The addressable market for AI-server MLCCs is projected at ~$900 mn by 2027; if AI infrastructure demand keeps beating expectations, it could top $1 bn by 2030.
Inventory is thin: after two years of destocking, distributor inventory averages ~1.5 months and downstream customers hold less than one month. Lead times for high-end products already exceed 20 weeks. Morgan Stanley expects tightness to persist through H2 2026 and into 2027.
Stocks have already surged — does the thesis still hold?
The MLCC sector is up 434% YTD; SEMCO alone has rallied 763%. The sector's price-to-book has hit 9.6×, far above its historical average of 1.7×.
This means → the investment case has shifted from "buy undervalued cyclicals" to "pay a premium for scarcity and accelerating earnings revisions."
Morgan Stanley flags two key checkpoints: whether contract prices can rise further, and whether MLCC content for the Rubin / VR200 platform is ultimately confirmed. These signals will determine if the current structural premium can keep delivering.
Content is for reference only, not financial advice.