The accelerated buildout of AI servers is shifting pressure onto much smaller electronic components than GPUs, but present in the thousands within each system. High-performance multilayer ceramic capacitors (MLCCs) currently have lead times that, for certain products, reach about 30-40 weeks, while conventional models maintain notably more stable availability. This difference indicates a transformation in passive component demand linked to AI data centers.
The key points of MLCC shortages in 20 seconds
- Conventional MLCCs generally have much shorter lead times compared to advanced models.
- Some high-capacity capacitors are facing 30-40 weeks of wait, depending on supplier and reference.
- An AI server may require several times more MLCCs than a conventional one.
- Murata, Samsung Electro-Mechanics, and Taiyo Yuden are prioritizing higher-value products.
- This pressure is already leading to price increases and long-term supply contracts.
The situation again illustrates how AI infrastructure depends on a supply chain far beyond NVIDIA, AMD, or HBM memory. An accelerator consuming hundreds or even over 1,000 watts needs a complex power network to keep voltages stable as load changes rapidly.
That’s where MLCCs come in.
These tiny and relatively unknown components outside the electronics industry store small amounts of energy, filter electrical noise, and help stabilize power supplies to processors, memories, and other circuits.
An AI server may contain thousands of them.
With AI, the number and, especially, the required performance levels are increasing.
From 14 weeks to nearly a year: not all MLCCs face the same issue
Talking about a “MLCC shortage” can lead to incorrect conclusions. There is currently not the same level of tension across the entire market.
Differences depend on capacity, size, voltage rating, operating temperature, reliability, and application.
Data from distributors collected by Asian media in August show many conventional MLCCs with lead times of about 14-18 weeks. High-capacity and high-voltage models can be around 15-20 weeks.
The jump appears mainly in certain high-performance references.
Some Samsung Electro-Mechanics MLCCs have reached approximately 40 weeks, while some high-capacity Murata products hit around 30 weeks in July, compared to about 24 weeks a month earlier.
These figures vary widely among references and distribution channels, so they cannot be extrapolated to the entire production of both manufacturers.
However, the overall picture aligns with other market indicators.
TrendForce reports that the book-to-bill ratios, comparing new orders to shipped products, reached at the end of June levels of 1.30 for Murata, 1.31 for Samsung Electro-Mechanics, and 1.25 for Taiyo Yuden, the highest since the pandemic. A value above one indicates that orders are coming in faster than products are being shipped.
| Type of MLCC | Observed lead times in 2026 | Situation |
|---|---|---|
| Conventional models | ~14-18 weeks | Relatively stable |
| High capacity / high voltage | ~15-20 weeks | Increased pressure |
| High-end for servers | More than 20 weeks | Tight supply |
| Some Murata references | Up to ~30-36 weeks | Limited availability |
| Some Samsung Electro-Mechanics references | Around ~40 weeks | Particularly tight cases |
Lead times are indicative and depend on reference, volume, manufacturer, contract, and distributor. They do not mean that all MLCCs from a manufacturer will have these wait times.
Why an AI server needs so many capacitors
The explanation lies in electricity.
An AI GPU or ASIC can change its power consumption extremely quickly depending on the workload. The power network must respond to these variations by maintaining voltage within very narrow margins.
MLCCs act as tiny energy reservoirs located near the chips, forming part of decoupling and filtering networks.
As consumption and computational density increase, this task becomes more demanding.
Murata acknowledges this trend in its industrial planning. The company expects demand for MLCCs in AI-equipped servers to multiply roughly fivefold between 2023 and 2030, especially requiring compact components, high capacity, and capable of operating at higher temperatures and voltages.
Some industry estimates suggest that an AI server consumes between five and thirteen times more MLCCs than certain conventional configurations, though the exact figure depends heavily on the specific server and architecture compared.
What matters is not just quantity.
AI systems specifically need the most difficult-to-manufacture references: small, high-capacity, thermally resistant, and highly reliable components.
These products use more industrial capacity per unit than simple consumer electronics ranges.
Production is shifting toward components for AI
This difference explains another consequence.
Manufacturers cannot instantly expand capacity, so they must decide which products to produce with existing lines.
MLCCs for AI servers also offer higher added value. Leading Japanese and South Korean manufacturers are gradually allocating more capacity to these ranges.
TrendForce expects manufacturing to be heavily occupied with AI-related orders during the second half of 2026, coinciding with the release of new platforms from NVIDIA, Google, and AMD. In July, the firm warned that this could extend lead times and cause further price increases.
This may also indirectly impact the rest of the market.
If factories dedicate more lines to advanced MLCCs, less capacity remains for conventional products, leading to potential shifts of orders to Taiwanese and Chinese suppliers.
The bottleneck begins in high-end components and can cascade downward.
Manufacturers are recovering capacity to raise prices
Prices already reflect this shift.
Samsung Electro-Mechanics announced a 30% price increase for certain products under its Markup Business Code starting 08/01/2026, according to TrendForce. Taiyo Yuden has also requested new adjustments, warning that even with these, meeting certain delivery deadlines might be difficult.
This situation is quite different from periods of inventory excess, when manufacturers and distributors competed aggressively on price.
Now, customers are seeking supply assurance.
In June and July, Samsung Electro-Mechanics closed two MLCC contracts with major international clients worth 453.99 billion and 295.12 billion won, respectively. Industry analysts in South Korea estimate that, converted to equivalent conventional product capacity, these agreements could account for about 10% of the company’s expected ceramic production in 2027.
Long-term agreements enable manufacturers to plan investments and give buyers greater security over future availability.
They also shift bargaining power.
Building a factory in a few months isn’t the answer
Increasing production of advanced MLCCs is not simply a matter of installing more machinery.
Manufacturing smaller, higher-capacity components requires precise control of extremely thin ceramic layers, electrodes, materials, thermal processes, and yields.
Manufacturers are expanding capacity, but investments take time to translate into market-ready products.
Murata, for example, has planned capacity increases to meet the expected growth in data center and AI applications. Its strategy includes continuous expansion and new investments to prepare for future demand.
Therefore, it is reasonable to expect some references will remain tight throughout the second half of 2026 and into 2027, though not all MLCCs will face shortages during that period.
Polarization is precisely the key issue.
Another tiny component that can limit massive data centers
The AI industry is constantly discovering components whose availability seemed secondary when volumes were lower.
First were GPUs. Then HBM memory and advanced packaging capabilities. Pressure has also reached PCBs, specialized laminates, fiberglass, electrical systems, transformers, and cooling solutions.
MLCCs add another piece to this puzzle.
A single capacitor costs vastly less than a high-end GPU, but a server requires thousands. Replacing one reference with another isn’t straightforward; capacity, voltage, temperature, size, and electrical characteristics must all match.
This shift can also economically benefit manufacturers. Combining high utilization factories, higher-value products, lower price reduction pressure, and longer contracts can improve average prices and margins.
Perhaps the most telling sign for 2026 is that the MLCC market is no longer moving uniformly.
While conventional components may remain available, the models needed for AI infrastructure are seeing extended orders that stretch over months.
As the power of GPUs and ASICs increases, these small capacitors are becoming more significant in the planning of some of the world’s most expensive servers.
Frequently Asked Questions
What is an MLCC?
An MLCC (Multilayer Ceramic Capacitor) is a multilayer ceramic capacitor. It’s used to store small amounts of energy, stabilize voltages, and filter electrical noise in virtually any electronic device.
Why do AI servers need more MLCCs?
AI GPUs and ASICs operate with very high power and rapid load changes. They require power networks with numerous high-capacity capacitors located close to the chips to maintain stable voltage and filter disturbances.
How long does it currently take to deliver an MLCC?
It depends entirely on the product. Many conventional references are approximately 14 to 20 weeks, while certain advanced ranges have reached 30-40 weeks in 2026.
Will supply issues persist into 2027?
Expanding capacity takes time, and AI infrastructure demand continues to grow. Sector analyses suggest some advanced ranges may remain tight into 2027, though the situation will vary across different MLCC types.
