A Market Split in Two

Murata and Samsung Electro-Mechanics (SEMCO) opened July with price increases of 10% to 40% on compute and high-end automotive-grade capacitors. But the headline number hides a starkly polarized market:

  • High-capacitance server-grade MLCCs (10uF, 20uF) are seeing spot price jumps of 50% to 60%, with specialized models doubling.
  • General-purpose MLCCs remain stable at 8-to-12-week lead times.

In other words, the pain is concentrated exactly where AI hardware needs it most.

Why AI Racks Are Driving It

Morgan Stanley and Goldman Sachs both flag a severe structural mismatch: next-generation AI platforms require up to a 182% increase in MLCC content value per rack. That single figure explains the supercycle - consumer and automotive brands are now aggressively competing with hyperscalers for the same high-capacitance capacity, and Murata's portfolio-wide hikes are the result.

The Tantalum Knock-On Effect

The squeeze is not staying inside the ceramic aisle. A tantalum capacitor shortage - compounded by Panasonic price hikes of 15% to 30% - is forcing engineering teams to redesign power-management sub-circuits toward hybrid "tantalum + MLCC" configurations. That shift is, in turn, pulling even more demand into the already-short high-capacitance MLCC pool.

Design Around the Constraint

Buyers who treat this as a pure purchasing problem will lose. The teams staying ahead are working with engineering to derate and re-spec: substituting constrained high-CV values where the design allows, and transitioning to soft-termination MLCCs to prevent board-bending and thermal-cycling flex cracks on high-stress boards.

What Buyers Should Do Now

  1. Next 48 hours: Scrub BOMs for exposure to 10uF/20uF server-grade MLCCs and flag any single-sourced high-CV parts.
  2. Next 30 days: Partner with engineering to qualify soft-termination and smaller high-spec automotive MLCCs (such as Murata's latest) as design-approved alternates.
  3. Next 90 days: Lock rolling capacity agreements for high-capacitance passives so AI-server demand cannot cannibalize your allocation.