MLCC Lead Times Split Wide Open: High-End AI Server Parts Now at 5–10 Months While Commodity Stays Loose

By Procurement Priya · icboms supply-chain desk · data through August 21, 2026

A procurement manager at an ODM in the Pearl River Delta received two quotes from the same franchised distributor last month. The 100nF X5R 0805 — the commodity ceramic capacitor every BOM has at least a dozen of — was available in four weeks. The 22µF X6S 1206 directly above it on the same power rail: twenty weeks, and climbing. "We have not seen that kind of divergence since 2018," the manager told icboms. She was describing exactly what three independent research firms — TrendForce, Astute Group, and New Electronics — confirmed in the past four days: the MLCC market is splitting in two, and the divide runs along the line separating standard parts from those demanded by AI server power architectures.

This article explains what is driving the divergence, which specific part categories are tight, what the material cost environment looks like upstream, and what buyers should be doing today with their high-capacitance capacitor BOM.

The market has split — and the split is widening

The clearest picture comes from TrendForce, which on August 18, 2026 published its latest MLCC market monitor. The headline finding: high-end MLCC lead times have stretched to a range of five to ten months, while standard commodity parts remain available on lead times closer to their pre-shortage norms. The research firm added that the tightness in high-end parts may not ease until well into 2027 — not because demand is fading, but because capacity additions at the large MLCC houses are landing in the general-capacitance segment, not the high-cap/high-voltage grades required by AI server voltage-regulator modules, bus termination, and decoupling planes.

Astute Group, in an analysis published August 21, went further on the demand side: AI server equipment is absorbing MLCC production capacity at a rate that has not been seen since the 2018 shortage cycle — the last time the industry declared a structural MLCC crisis. Order books at several distributors are at those same elevated levels. The implication for buyers is immediate: if your BOM includes high-capacitance parts above 10µF in 1206 or larger case sizes, your standard distributor may already be quoting you a lead time that does not match the lead time you got six months ago.

New Electronics confirmed the same pattern on August 19, 2026: lead times for high-capacitance multilayer ceramics are diverging from standard parts in a way that is becoming structurally significant. The divergence is not a glitch — it reflects a genuine mismatch between where fab capacity is being added and where demand is concentrating.

Why AI servers swallow a different kind of MLCC

A standard server or networking switch uses a mix of MLCCs in the 1µF to 22µF range across many voltage rails. An AI server — specifically the GPU accelerator blades and the OAM modules populating NVIDIA HGX or similar racks — has a fundamentally denser power architecture. High-bandwidth memory, the compute GPU, and the NVLink switch all require clean, low-noise power at multiple rails operating at 0.8V, 1.2V, 1.8V, and 3.3V simultaneously. Each rail demands bulk decoupling capacitance in the 22µF to 220µF range at the appropriate voltage rating — and those are exactly the parts under pressure.

Nichidenbo, a Taiwanese passive component distributor that serves as a bellwether for the MLCC spot and contract market, reported second-quarter fiscal 2026 results on August 11, 2026. Net profit surged 643% year-on-year; EPS hit a record NT$3.21. Management attributed the gain directly to sequential MLCC shipment growth driven by AI server demand. When a distributor that moves millions of capacitors per quarter is posting 643% profit growth on the back of AI server orders, the demand signal is not speculative — it is already in the supply chain.

Complementing this, All About Circuits on August 19, 2026 reported that Murata, TDK, and Knowles — three of the five largest global MLCC houses — each raised their internal production expectations and introduced expanded product lines, a move that typically precedes or accompanies periods of tight supply for certain product categories. The simultaneous action by three major manufacturers within days of each other is unusual and suggests a shared view that near-term supply will not fully satisfy demand.

On the material side, Sinocera, a major Chinese zirconia powder producer whose output feeds into MLCC dielectric formulations, raised zirconia powder prices by approximately 40% in early August 2026, as reported by finance.biggo.com on August 9. Zirconia is a key raw material in the manufacture of high-capacitance barium-titanate-based MLCC dielectrics. A 40% increase in a critical input cost does not show up immediately in spot pricing — it works its way through inventory cycles and manufacturer cost bases before it reaches buyers as a formal price letter. But it is a strong leading indicator that the cost environment supporting high-cap MLCC prices will remain elevated through the rest of 2026 and into 2027.

What we carry — Kyocera AVX and Samsung Electro-Mechanics

We carry high-capacitance parts from two manufacturers whose catalog lines map directly to the AI server bottleneck:

Kyocera AVX — Tantalum Polymer, TPSE series (150µF to 680µF, 4V to 35V):

Kyocera AVX is one of the two dominant manufacturers of polymer tantalum capacitors — a technology that competes directly with high-cap MLCCs in applications where ripple current capability and ESR stability matter more than volumetric efficiency alone. Their TPSE and TPSD series are widely used in voltage regulation modules for server boards, particularly in input and output bulk capacitance positions where the AI server's high current transients require a capacitor that can handle large charge/discharge cycles without degradation.

TPSD157K016H0125 — 150µF, 16V, K-density (low ESR), D-case (EIA 7343-31), tantalum polymer. On our shelf. MOQ: 10 units. Available for sample and production RFQ.
TPSE686M020R0150V — 68µF, 20V, M-density, E-case (EIA 7343-20). For lower-voltage rails in server power designs. Available.

The Kyocera AVX parts above represent the types of high-capacitance, high-voltage capacitors used in AI server power architecture. If your design uses a VRM with bulk input capacitance above 22µF at 16V or higher, the Kyocera AVX polymer tantalum series is a proven alternative to the equivalent high-cap MLCC — and may have better availability in the current environment.

Samsung Electro-Mechanics — High-Cap MLCCs (100nF to 10µF, 4V to 50V, 0201 to 1812):

Samsung EM is the second-largest MLCC manufacturer globally and a primary supplier of the high-capacitance parts that are under the most pressure in AI server applications. We carry a broad range of Samsung EM CL series parts scoring 93 to 95 in our catalog, including:

CL31B104KACNBNC — 100nF, X7R, 25V, 1206. One of the most widely used bypass-capacitance values in server board designs. On our shelf; production RFQ available.
CL31C331KBCNBNC — 330pF, C0G/NP0, 50V, 1206. For high-frequency decoupling positions where temperature stability is critical.
CL21B104MOCNBNC — 100nF, X5R, 16V, 0805. Standard bypass cap for lower-voltage rails.

Taiyo Yuden — Inductors and Chip Components:

Taiyo Yuden — a top-five global MLCC maker — is also a major supplier of power inductors and chip beads used alongside capacitors in AI server PDN (power delivery network) designs. Several Taiyo Yuden inductor series are available through our catalog:

NR3015T470M — 47µH shielded wire-wound inductor, 1.2A saturation current. For DC-DC converter input/output filtering in server power modules.
NR4018T3R3M — 3.3µH inductor, 2.5A, for high-frequency power rail conditioning.

If your BOM is tight on high-capacitance MLCCs, Taiyo Yuden inductors used in conjunction with lower-capacitance ceramics in active filtering configurations may offer a partial workaround — though this requires board-level re-evaluation.

How to buy in this market — a six-point plan

Separate your BOM into two baskets immediately. High-cap (above 10µF, 1206 and larger, rated 6.3V and above) goes into the "AI server tight" basket; everything else stays in the "standard" basket. Quote them as separate line items. Do not wait for a single quote to reveal two different lead times.
Lock your high-cap requirements for 2027 now. The TrendForce data suggests tightness may persist into 2027. If you have a rolling 12-month forecast, convert the first two quarters of 2027 requirements from forecast to firm purchase orders on your franchised distributor — especially for Kyocera AVX TPSD/TPSE and Samsung EM CL31B/CL31F series.
Evaluate polymer tantalum as a direct substitute where ESR and ripple current matter. The Kyocera AVX TPSD157K016R0125V at 150µF/16V can substitute for an equivalent high-cap MLCC in many VRM bulk positions. Verify the landing pattern and voltage derating with your hardware team before committing to a substitution, but do the evaluation now rather than when the MLCC lead time hits twenty weeks.
Watch the zirconia powder price signal. Sinocera raised prices 40% in August 2026. That cost increase will reach component pricing over the next one to three quarters. Buyers who locked pricing in Q3 2026 will have a window before the next round of component price letters. Use that window.
Watch Nichidenbo's monthly revenue prints. A distribution business with 643% net profit growth in a single quarter is not a niche player — it is a proxy for the underlying spot market. Declining sequential growth at Nichidenbo would be an early signal that high-cap MLCC demand is cooling. Flat or rising sequential growth means the tightness continues.
Do not chase spot market offers for high-cap parts below distributor list price. During the 2018 shortage cycle, brokers and independent brokers entered the market with "excess inventory" that turned out to be mis-marked date codes, dry-packed parts that had exceeded their moisture shelf life, or outright counterfeits. If a quote for a 100µF 25V 1206 X5R comes in significantly below market, verify the lot traceability and request a certificate of conformance before assembly.

Data notes

Data cutoff: August 21, 2026. Sources: TrendForce (August 18, 2026 — MLCC lead times five to ten months, tightness into 2027); Astute Group (August 21, 2026 — order books at 2018 levels); New Electronics (August 19, 2026 — lead time divergence); All About Circuits (August 19, 2026 — Murata, TDK, Knowles raise production expectations); finance.biggo.com / Nichidenbo earnings (August 11, 2026 — 643% net profit growth, AI server MLCC shipment growth); finance.biggo.com / Sinocera zirconia (August 9, 2026 — zirconia powder price +40%). All data cited as directional planning information, not guarantees. Verify current lead times and spot availability with your distributor before placing orders. MOQ varies by part and distributor; contact icboms for specific quotes on the Kyocera AVX and Samsung EM lines above.

Need a quote on high-capacitance capacitors for an AI server BOM? Send us your target MPNs, quantities, and required delivery date. We respond within one business day with cross-reference options, pricing, and lead time for the Kyocera AVX TPSD/TPSE series, Samsung EM CL series, and Taiyo Yuden inductor and MLCC lines in our catalog.