If you work in hardware procurement, you probably remember the morning of March 17 clearly. That’s when Murata — the company responsible for roughly one in three MLCCs shipped worldwide — told customers their prices were going up. Not by a little. By 15 to 35 percent, effective April 1.
The natural reaction was: we’ve seen this before. MLCCs had a shortage in 2017–2018, prices spiked, then normalized. Cyclical. Wait it out.
I don’t think that’s the right read this time. And the reason comes down to a single number that changed everything about MLCC demand.
The Number: 30,000
That’s how many MLCCs go into a single NVIDIA GB300 compute unit. Thirty thousand ceramic capacitors per system.
For context: a standard 1U server uses about 2,200 MLCCs. An H100-based server board uses roughly 6,500. The GB200 compute node needs 12,000–15,000. And the GB300 — the backbone of next-generation AI training clusters — requires 30,000.
Now multiply that by scale. A full NVL72 rack packs 441,000 MLCCs. Hyperscalers are deploying thousands of these racks. NVIDIA alone is consuming MLCC capacity equivalent to what previously served entire consumer electronics segments.
Murata has publicly acknowledged that AI server MLCC orders currently exceed double their production capacity for high-end products. That’s not a forecast. That’s existing backlog.
Why “Wait It Out” Won’t Work This Time
The 2017–2018 MLCC shortage was driven by simultaneous demand across many segments — smartphones, automotive, industrial. When smartphone growth cooled, supply caught up. Classic cycle.
This time is structurally different for three reasons:
First, the demand is concentrated in products that require the most advanced MLCCs. AI servers don’t use commodity 0402 capacitors. They need high-capacitance, high-voltage MLCCs for power delivery network decoupling — exactly the products that are hardest to manufacture and have the longest capacity expansion timelines.
Second, AI infrastructure investment is not cyclical. Unlike cryptocurrency mining (which boomed and busted), AI compute demand is backed by every major tech company’s multi-year capex commitments. Microsoft, Google, Meta, Amazon, and dozens of Chinese hyperscalers are all building simultaneously. And inference demand — which grows as more AI models deploy to production — adds a compounding layer on top of training demand.
Third, expanding high-end MLCC capacity takes 18–24 months. These aren’t commodity products you make on any production line. High-capacitance MLCCs require dielectric layers thinner than 0.5 µm, stacked 1,000+ layers deep with near-zero defects, using proprietary ceramic formulations. You can’t shortcut the physics of building a new production line.
Put these together and the conclusion is uncomfortable but clear: this is a permanent demand plateau shift, not a temporary spike. Prices may ease modestly as new capacity comes online in late 2027, but they are unlikely to return to 2025 levels.
The Domino Effect
Within weeks of Murata’s announcement, the rest of the industry followed:
Samsung Electro-Mechanics raised prices 15–20% in May. Taiyo Yuden went 15–25% in Q2. Yageo (through Kemet) implemented 10–15% increases across their broad portfolio. TDK is negotiating adjustments as this goes to print.
When the market leader with 30% share moves, followers are mathematically inevitable. Remaining manufacturers cannot absorb the demand overflow without repricing. This is now an industry-wide event, not a single-supplier issue.
The Good News: It’s Not Everything
Here’s what’s important to understand — and what some breathless industry coverage misses:
Standard consumer-grade MLCCs are fine. If your BOM consists primarily of 0402/0603 capacitors at 100nF–10µF in the 6.3V–25V range, commercial temperature grade, you are largely unaffected. Supply is adequate. Pricing is stable.
The shortage and price pressure is specifically concentrated in:
– High-capacitance MLCCs (47µF+ in small packages)
– High-voltage ratings (50V+) in automotive temperature ranges
– AEC-Q200 qualified products
– Server-grade MLCCs optimized for power delivery networks
If none of these describe your BOM, exhale. If even one does, keep reading.
What To Do Right Now
This week: Audit your BOM for exposure. Identify every MLCC line item that falls into the high-end category — automotive grade, high-CV, server grade. Calculate what a 20% price increase does to your unit cost. If the answer is “tolerable,” you have time to be strategic. If the answer is “margin-destroying,” act fast.
This month: If you have existing contracts at pre-increase pricing, maximize your order volumes within contract terms. Every MLCC purchased at old pricing is direct margin protection. Also: check Samsung Electro-Mechanics and Yageo cross-references. The same capacitance, voltage, and package from an alternative manufacturer may still be available at a lower premium — for now.
This quarter: Begin qualifying second sources for your highest-volume high-end MLCC part numbers. This isn’t about replacing Murata permanently. It’s about having options when allocation tightens further. If you’ve never used GigaDevice or Fenghua for any MLCC specifications, now is the time to evaluate whether domestic Chinese alternatives could serve any of your non-automotive needs.
For your next product: Consider whether your power delivery network design can tolerate slightly larger package sizes. A 1210 in the same capacitance and voltage is typically less supply-constrained than an 0805 or 0603. One well-chosen advanced MLCC (like Murata’s GCM21BE71H106KE02 — an 0805, 10µF/50V part mass-produced since mid-2025) can replace two or three standard capacitors. Fewer unique part numbers means less supply chain vulnerability.
The Strategic Question You Need to Answer
Here’s the uncomfortable budget conversation that procurement leaders need to have with their finance teams:
Should we budget for MLCC prices 15–30% above 2025 levels as our new baseline?
My answer is yes. If prices come down — great, you’ve gained margin. If you budget at old prices and they don’t come down, you face either margin compression or the impossible choice between accepting extended lead times and paying spot-market premiums of 50–200%.
The asymmetry of outcomes favors conservative budgeting. Hope for the best, budget for the structural shift.
A Final Thought
I’ve been sourcing components long enough to know that price increases feel like a crisis in the moment but become the new normal within two quarters. Teams that adapt their budgets and diversify their supply base early don’t just survive — they gain a competitive advantage over competitors still hoping for a return to 2024 pricing.
This is one of those moments. Adapt early.
Facing MLCC allocation challenges? Cosolvic maintains sourcing channels across Japanese, Korean, and domestic manufacturers for high-end capacitors. Submit your BOM for current availability and pricing.