Multilayer ceramic capacitors (MLCCs) — the "rice of the electronics industry" — are experiencing an unprecedented supply-demand imbalance.
Demand: Explosive Growth in AI Server Consumption
The latest global MLCC supply/demand model from Morgan Stanley paints a staggering picture:
• Standard server — MLCC per unit: 1,800-2,500 pcs.
• 8-GPU AI server — per unit: ~20,000 pcs. (~2,500 pcs. per GPU)
• NVIDIA GB200 — per board: ~6,500 pcs.
• NVIDIA Rubin — per board: ~12,000 pcs. (nearly double the GB200)
• MLCC cost per rack: from $3,000 (H100) → $12,000 (GB200) → $22,000 (Rubin VR200)
Supply: Capacity Expansion Lags Far Behind Demand
The MLCC capacity expansion cycle is 18-24 months (from construction to capacity ramp-up). The AI server demand explosion took just 12 months. Samsung Electro-Mechanics is discussing with a major US CSP an MLCC supply contract for AI servers worth ~500 billion KRW (~2.2 billion CNY), but new capacity will not arrive before Q3 2027. Japanese manufacturers (Murata, TDK) are actively shrinking mid- and low-tier segments, concentrating on automotive-grade and AI server MLCCs.
Shortage Forecast
MS model: MLCC shortage widens from Q2 2026, persists through all of 2027, and by 2028 the gap remains at ~6%. ASP (average selling price) to rise >50%. The scale of the shortage may surpass the 2017-2018 super-cycle. For small and medium electronics manufacturers, MLCC procurement strategy for 2026-2028 must shift from "just-in-time buying" to "strategic inventory building" with active evaluation of domestic alternatives.
Import Substitution Window
Japanese and Korean capacity is shifting toward the upper MLCC segment. Mid- and low-tier orders are accelerating their shift to Chinese manufacturers. Fenghua Advanced Technology, Sanhuan Group, and Weirong Technology have significantly increased capacity utilization in 2026, expanding their product lines from consumer-grade to industrial and automotive-grade.
Comments …