MLCC, the Trigger of the Next AI Semiconductor Supercycle

The AI semiconductor industry has now gone through three major bottlenecks. First was the GPU shortage, followed by HBM supply constraints, and CoWoS advanced packaging capacity emerged as the third constraint. And in early 2026, a fourth bottleneck is quietly emerging — the passive component industry, led by MLCCs.

Murata President Nakajima said last February that "customer inquiries for AI MLCCs are currently double our production capacity." NVIDIA's next-generation server platform VR200 NVL72 is estimated to be equipped with approximately 600,000 MLCCs, an increase of more than 30% compared to the existing GB300. It's not just a simple increase in demand — it's a structural shift.

Why MLCC? AI accelerators generate instantaneous current fluctuations during high-load calculations. MLCCs play a key role in stabilizing this voltage, suppressing noise, and compensating for instantaneous current drops. As power modules evolve from 10kW to 15kW, the number of MLCCs installed per system has surged from 2,200 to over 20,000, and is expected to exceed 30,000 in next-generation platforms. MLCCs have already risen to the 3rd highest cost item in the AI server BOM, following GPUs and memory (compiled by Nichidenbo).

The historical context is clear. During the MLCC supercycle in 2018, Samsung Electro-Mechanics' MLCC division's operating profit margin soared to a peak of 42%. Currently, Murata's margin already exceeds 30%, while Samsung Electro-Mechanics remains at 11.7%. This gap is the potential for stock price appreciation over the next 12-24 months.

The market has a stark dual structure. Tier-1 MLCCs from Japan and Korea (Murata, Samsung Electro-Mechanics) monopolize AI server demand, while Taiwanese and Chinese companies are struggling with capacity utilization rates of 60-70% amid sluggish consumer electronics demand and price competition. This dichotomy is not a short-term phenomenon but a structural one — the ultra-thin ceramic technology and reliability certifications required for advanced MLCC manufacturing cannot be replicated in the short term.

The structural characteristics of MLCC — oligopolistic market, technological barriers to entry, several-fold increase in usage due to AI demand — are observed in multiple similar sectors in the AI semiconductor industry map. Investors must recognize this pattern and read the entire value chain.

Advanced packaging (CoWoS) is currently the hottest bottleneck. TSMC's CoWoS-L recorded 1,000% YoY growth in 2025, yet NVIDIA preempted 60% of the available capacity. Glass Substrate is emerging as a next-generation substrate material and is highly likely to become the 'next MLCC shortage'. Optical interconnects (CPO/OIO) are a new bottleneck emerging from the physical limitations of electrical signal transmission — Ayar Labs ($370M), Lightmatter ($400M, $4.4B valuation) are here. Semiconductor testing equipment overwhelmed almost all semiconductor categories with 48.1% growth in 2025.