Market news |
By Nick Flaherty
The rapid advancement of AI is driving a shift in the delivery of key components, particularly MLCC ceramic capacitors.
The adoption of 48V power lines is accelerating, as their higher voltage compared to 12V systems reduces power loss and boosts board design efficiency. At the same time, resonant circuits, such as resonant converters (LLC), are being used more and more for high-efficiency power conversion. The introduction of new 800V power systems marks the next step in AI server power architecture, with AI power systems becoming increasingly complex and sophisticated.
This is driving a shift in a key passive component, the multi layer ceramic capacitor (MLCC). The production of MLCCs is dominated by Japanese companies – Murata, TDK, Kyocera AVX and Taiyo Yuden – as well as Samsung Electro-Mechanical in Korea and Yageo in Taiwan.
The AI wave is shifting from semiconductors to passive component MLCCs, says Samsung. AI servers use ten to 15 times the number of MLCCs in comparison to general-purpose servers (see below). Not only are they driving volume, but also expanding demands into ultra high capacitance and high voltage MLCCs which require advanced technologies.
As a result major Japanese and South Korean suppliers have shifted their production capacity toward components for AI applications. This, in turn, is steadily constraining the supply flexibility of consumer MLCCs quarter by quarter.
Market analysts and distributors are warning that high-capacitance MLCCs are entering a period of shortages as tier-one manufacturers divert production lines to support the rapid expansion of AI infrastructure.
The situation is reminiscent of 2021 when supply of MLCCs became a critical issue for equipment makers in the Covid shortage.
The latest market research from TrendForce in Taiwan shows that the MLCC market in 2Q26 is showing a clear split between robust AI-driven demand and soft consumer demand. The Iran conflict has pushed up oil and natural gas prices, driving higher energy and transportation costs while inflation is expected to intensify and weigh on end-market demand and corporate capital spending.
These effects are gradually feeding through to the electronic components supply chain. Meanwhile, rising prices for key metals such as silver, aluminium, and copper have led to price increases of 10 to 15% on average for passive components.
The tight supply of key components has prompted OEMs such as Dell and HP to adopt strategic inventory-building measures. Some low- to mid-tier notebook orders originally scheduled for 3Q have been pulled forward into 2Q production and shipment to lock in lower pricing expectations, stimulate demand, and boost revenue.
In response to shrinking production capacity and strict inventory controls for consumer MLCCs, agents based in Taiwan and Mainland China have begun pre-emptively stocking X5R standard products (capacitance values between 1000pF and 10µF).
In April 2026, Taiyo Yuden took the lead by raising prices for low-capacitance consumer and automotive MLCCs by 6% to 13%. By early May, some equipment makers had already completed their 3Q26 price negotiations with MLCC suppliers, with process recovering. As the majority of ODMs begin a new round of price negotiations in late May, a key question is whether current market conditions drive a rebound in MLCC prices for automotive and consumer applications.
Mordor Intelligence values the Low Voltage MLCC market at US$6.17 billion in 2025 and estimated to grow from $18.97 billion in 2026 to $42.18 billion by 2031, a significant CAGR of 17.33% for a passive component.
Demand in 5G handsets, AI servers and electrified vehicles require ultra-compact, low-loss capacitors that can be mounted in ever-tighter spaces while delivering higher capacitance per volume. While Asia-Pacific retains its manufacturing and consumption lead, near-shoring and semiconductor stimulus programs are pushing new capacity closer to end customers in North America.
The leading manufacturers are expanding production of layers less than 0.3 µm thick to further reduce the size of the MLCCs without sacrificing reliability. Meanwhile, price fluctuations in nickel and palladium complicate cost forecasting, prompting recycling initiatives and innovations in materials.
MLCC market forecast from Mordor Intelligence
This demand has already been reflected in recent shipments and revenue performance of contract equipment makers such as Quanta, Wistron, and Compal. However, since OEMs have not revised their full-year shipment forecasts upward, risks are building for a weaker-than-usual peak season in the second half 26, says TrendForce, along with potential order corrections.
Strong demand from AI servers has prompted Japanese and Korean manufacturers to reallocate capacity from consumer-grade products to high-end MLCCs. The industry book-to-bill (BB) ratio improved from 0.89 in March to 0.92 in April, while the leading suppliers have maintained ratios consistently above 1, indicating expansion
Taiyo Yuden has already raised prices for low- to mid-capacitance consumer MLCCs and select automotive products by approximately 6 to 13% for distributors in China. Meanwhile, Yageo and Walsin Technology Corporation (WTC) are negotiating price adjustments on a case-by-case basis for certain loss-making products, without announcing broad price hikes. Industry leaders Murata and Samsung Electro-Mechanics have yet to make formal announcements, but overall pricing sentiment is shifting from cautious observation to tentative upward adjustments.
TrendForce expects that AI server projects that are ramping up at the end of the year will tighten supply of high-end MLCCs and drive up prices.
However, inventory corrections from the PC and notebook market, along with geopolitical uncertainties and shifts in monetary policy, will remain key risks for consumer MLCC demand and pricing.
This is backed up by UK distributor Astute, which sees rising passive component lead times threatening non-AI hardware margins.
The imbalance stems from the surge in AI server deployments, which require significantly higher concentrations of specialised passives compared to standard enterprise hardware. It points to analysis by Evertiq that Murata and Taiyo Yuden are prioritising high-performance, high-reliability components for data centres. This reallocation of capacity is causing lead times to stretch for high-capacitance parts in the 1206 and 1210 case sizes, typically used in power management stages.
While standard 0402 and 0603 MLCCs remain readily available, larger devices are seeing lead times extend beyond 20 weeks. This volatility complicates manufacturing schedules for industrial automation and medical electronics, where redesigning a PCB to accommodate different capacitor footprints is costly and time-consuming.
Market analysis suggests that the shortage is not a result of raw material scarcity, but rather a deliberate strategy by manufacturers to maximise margins.
“Suppliers are hesitant to add new capacity for lower-margin components despite the clear signs of recovery,” said Evertiq. This disciplined approach to capital expenditure means that any sudden spike in demand for consumer or industrial electronics could result in broader stockouts across the bill of materials.
Inventory at distributors can provide a temporary buffer, but as tier-one OEMs consume available stock, smaller players face increased competition for remaining allocations. Sourcing strategies are moving away from just-in-time models toward long-term agreements for critical passive components to ensure production continuity through 2026.
“The divergence between AI-driven demand and general industrial recovery is creating a dangerous procurement gap for standard hardware manufacturers. Firms must secure allocations for high-capacitance MLCCs now, as lead times will likely deteriorate further when automotive demand aligns with the current server build-out,” said Damian Semple, Franchise Marketing Manager at Astute.
This is not a surprise. Back in March Murata opened its latest MLCC plant in Japan at Izumo. The €255m plant has been under construction since 2024 and provides up to 70,000 sq m of production. This is intended to help Murata to respond to the medium- to long-term increase in demand for MLCCs across Japan and China as the market continues to expand.
“By distributing our MLCC production bases across Japan, ASEAN, and China and expanding our production capacity in a balanced manner, we are working to establish a system that enables us to respond flexibly to changes in the market environment and to growing demand,” said the company.
High capacitance MLCCs are increasingly in demand to support both GPUs and CPUs that consume thousands of amperes of current at a low voltage of 0.8V. The total capacitance of the MLCC has to increase to ensure a stable power supply for GPU power Dynamics.
A Grace Blackwell GB200 server board from Nvidia requires approximately 6,500 MLCCs, while the coming Rubin architecture will increase this to roughly 12,000 devices per board. Simultaneously, cloud service providers (CSPs) such as Microsoft, AWS, Google, and Meta are steadily increasing orders for in-house ASIC chips and CoWoS advanced packaging, driving up long-term demand for high-end MLCCs.
MLCCs placed near high-performance computing board GPUs and CPUs play a decoupling role to mitigate rapid current changes. As the performance of chips increases, the mounting area is reduced, whilst the required capacitance grows. This requires capacitances of over 47µF within 0402 inch size or 10µF in 0603 inch size.
The capacitance of MLCCs which undergo surface mount process near GPU balls is also increasing. Embedded MLCCs and landside MLCCs that are mounted within the semiconductor package or right underneath reduce loop inductance significantly while increasing capacitance density. The power density per unit area is linked to the capacitance density, with X7T 0402 inch 22µF, X6S 47µF 2.5V MLCCs under active evaluation at Samsung
Another factor is power efficiency. While 48V systems require 100V MLCCs, a 120kW power delivery system for AC to 800V conversion will need larger 1kV–2kV MLCCs.
In response to GPU load current dynamics, large core power supply currents are necessary. VPD (Vertical Power Delivery) technology is a form of power module that not only shortens the power path as much as possible, but also increases power density.
Last year Murata Manufacturing started the world’s first mass production of the 0402 size MLCC that measures 1.0 × 0.5 mm with a capacitance of 47µF for AI systems.
Equipment in AI servers and data centre requires increasingly high component density, so optimized component placement within limited PCB areas is paramount. As a result, there is increasing demand for capacitors that offer both miniaturization and higher capacitance, along with high reliability under high-temperature conditions caused by heat generated from PCBs and integrated circuits (ICs).
Murata used its proprietary technologies in ceramic dielectric layers and internal electrode miniaturization for the production of the 47μF MLCC. Compared to Murata’s conventional 0603-inch size product with the same capacitance, the capacitor reduces mounting area by approximately 60%. It also delivers over twice the capacitance of Murata’s previous 22μF product in the same 0402-inch size.
The MLCC is available in two variants – the X5R (EIA) GRM158R60E476ME01 with an operating temperature range of -55 to +85°C, and the X6S (EIA) GRM158C80E476ME01 at -55 to +105°C. The ability to operate up to 105°C makes the X6S variant well-suited for placement near the GPU, and both devices feature a ±20% tolerance and rated voltage of 2.5Vdc.
Moving AI into smartphones is also driving demand for MLCCs. At the end of 2025 Kyocera started production of a 47µF capacitor in the 0402 form factor measuring 1.0mm x 0.5mm.
The Kyocera KGM05 MLCC
The 0402 is the most widely adopted size for smartphones and wearable devices and Kyocera has reduced the thickness of the dielectric and internal electrodes, expanding the capacitance per unit by approximately 2.1 times compared to the previous 22μF capacitor in the same size, reducing the number of components. Increased thermal resistance up to +105°C ensures high reliability in harsh temperature environments such as AI servers.
Taiyo Yuden in Japan has the third largest market share of MLCCs, and its latest embeddable MLCC achieves a capacitance of 22μF in a 1005 size (1.0 x 0.5 mm). This is designed for decoupling applications on IC power lines used in AI servers that require high precision in terms of flatness of the external electrodes for connection to the circuit.
For decoupling in AI power supply circuits, small, high-capacity MLCCs are required to handle large currents. Additionally, to minimize circuit loss and noise, it is important to route the power supply circuit close to the chips.
Mass production of the capacitor began at the Tamamura Plant (Sawa District, Gunma Prefecture) in August 2025 and samples are available for 20 yen per capacitor. The company says it is continuing to develop MLCCs with higher capacitance.
A US supplier has used a silicon process to develop higher capacitance embedded capacitors for these AI applications. The latest embedded silicon capacitors (ECAPs) from Empower Semiconductor deliver 9.34μF capacitance in a 2mm x 2mm package; 18.68μF in a 4mm x 2mm package; and 36.8μF capacitance in a 4mm x 4mm form factor.
“Our customers are under intense pressure to deliver greater performance with tighter power margins,” said Steve Hertog, Senior Vice President Worldwide Sales, Empower Semiconductor. “These ECAPs are a proven and practical way to deploy higher capacitance density into a smaller footprint right at the package level of the AI processor.”
The capacitors have low equivalent series inductance (ESL) and equivalent series resistance (ESR) while a wide bandwidth ultralow impedance delivers optimal power delivery network (PDN) performance and significantly improves overall power integrity. Each device has been engineered from a packaging perspective to meet the strict dimensional and tolerance requirements necessary for embedded deployment within AI and HPC processors.
www.murata.com; www.samsung.com; www.empowersemi.com; www.kyocera-avx.com; www.tdk.com; www.mordorintelligence.com; www.astute.co.uk; www.trendforce.com
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。