What Are MLCCs? Multilayer Ceramic Capacitors and the AI Buildout

25th Aug 2026

A plain-English guide to the parts the entire electronics industry runs on, and why Wall Street suddenly cares about components that cost fractions of a cent.

The market has spent three years obsessing over processors. GPUs, custom accelerators, high-bandwidth memory: the glamour components of the AI buildout are practically household names. But every one of those chips depends on a category of parts most investors have never thought about, made by companies many have never heard of, at prices often measured in fractions of a cent. In Asia’s electronics supply chain, they are so ubiquitous they carry a nickname: the rice of the electronics industry.

They are capacitors, and their most important modern form is the multilayer ceramic capacitor, or MLCC. In 2026, lead times for high-end MLCCs have stretched from roughly eight weeks to as long as five to ten months, the industry’s largest suppliers have pushed through their broadest price increases since 2018, and Goldman Sachs analysts have reportedly identified MLCCs as the third-largest cost item in an AI server, behind only the GPU and memory. Understanding why starts with understanding what a capacitor actually is.

What is a capacitor?

A capacitor is one of the three fundamental passive components in electronics, alongside resistors and inductors. In its simplest form, it is two conductive plates separated by a thin insulating layer called a dielectric. Apply a voltage across the plates and opposite charges accumulate on either side, storing energy in the electric field between them. Remove the source, and the capacitor releases that stored charge back into the circuit.

The comparison people usually reach for is a battery, but the two do very different jobs. A battery stores energy chemically and releases it gradually over minutes or hours. A capacitor stores energy electrostatically and can charge or discharge in microseconds or faster. If a battery is the fuel tank, a capacitor is a small pressurized reservoir sitting right next to the engine, absorbing surges and instantly covering shortfalls that the tank cannot respond to quickly enough.

Capacitance, the amount of charge a capacitor holds per volt, is measured in farads. A full farad is enormous by electronics standards; the parts populating circuit boards are rated in microfarads, nanofarads, and picofarads. What any given capacitor is good for depends on its capacitance, its voltage rating, how it behaves across temperature and frequency, and how fast it can move charge in and out.

What does a capacitor actually do?

Capacitors show up in nearly every job a circuit performs. In power supplies, they smooth the rippled output of converted AC into steady DC. In signal paths, paired with resistors and inductors, they form filters that pass some frequencies and block others, and they block DC while letting signals through. With a resistor, a charging capacitor becomes a timer, the basis of countless oscillators and clocks.

Their most economically important role today, however, is what engineers call power integrity: keeping the supply voltage rock-steady at chips that draw large, violently fluctuating amounts of current. Small capacitors placed immediately next to a processor, called decoupling or bypass capacitors, act as local reservoirs of charge. When the chip’s demand spikes for a nanosecond, the nearby capacitors deliver current before the main power supply can react, and they soak up electrical noise that would otherwise corrupt computation. Modern chips do not work without them. That single sentence explains most of what follows.

The capacitor family tree

Capacitors come in several major technologies, each defined by its dielectric material and each with a distinct role.

Ceramic capacitors, led by the MLCC, are the volume kings: tiny, cheap, non-polarized, and excellent at high frequencies. They dominate decoupling, filtering, and general-purpose duty across virtually every electronic product made.

Aluminum electrolytic capacitors deliver large amounts of capacitance inexpensively and handle bulk energy storage and smoothing in power supplies. They are physically larger, polarized, and have finite lifetimes, but nothing beats them for raw capacitance per dollar.

Tantalum and polymer capacitors pack stable, dense capacitance into compact, reliable packages, which earns them slots in servers, aerospace, medical devices, and other applications where space and dependability both matter.

Film capacitors use plastic film dielectrics and excel at high voltages and punishing conditions, making them staples of EV inverters, industrial power electronics, solar and grid equipment, and the front-end power stages of data centers.

Silicon capacitors are the newest branch: capacitors fabricated with semiconductor processes, so thin and electrically clean that they can sit inside or directly beneath advanced chip packages. In May 2026, Samsung Electro-Mechanics announced a silicon capacitor supply agreement reported at roughly $1 billion with a major global customer, a signal of how seriously the industry now takes package-level power delivery.

What is an MLCC?

The multilayer ceramic capacitor is the technology that made the modern electronics industry possible. Instead of two plates, an MLCC stacks hundreds of alternating layers of ceramic dielectric, commonly based on barium titanate, and metal electrodes, typically nickel. The stack is pressed and fired into a single monolithic chip with terminals on each end. Layering multiplies the effective plate area inside a package that can be smaller than a grain of sand, and advanced parts now use dielectric layers thinner than a micron.

The results are extraordinary on every axis that matters for mass production. A common 0402-size MLCC measures about 1.0 by 0.5 millimeters. Commodity parts sell for fractions of a cent. They have no polarity to get wrong, mount by the thousands via automated assembly, and perform superbly at the high frequencies modern chips operate at. Global MLCC output is measured in the trillions of units per year by common industry estimates, and third-party researchers such as Persistence Market Research size the global MLCC market at roughly $13 billion in 2026, with growth projected across automotive, industrial, communications, and computing end markets.

The counts inside everyday devices explain the volumes. Industry estimates put roughly 1,000 to 1,200 MLCCs in a modern smartphone. An electric vehicle typically carries 5,000 to 10,000, and premium software-defined EV platforms are trending well beyond that. A standard enterprise server contains roughly 2,500. For decades, those numbers made MLCCs a quiet, cyclical business tied to phone and PC demand. Then AI servers arrived.

Why AI changed the math

The defining feature of an AI accelerator, from an electrical standpoint, is that it is a power monster. Reported figures for flagship data center GPUs have climbed from roughly 700 watts per chip to well over 1,000, with next-generation platforms reportedly targeting substantially more. Those chips run at core voltages below one volt, which means the current flowing into a single processor can exceed a thousand amps, delivered into a die a few square inches in area, with a tolerance for voltage noise of only around plus or minus two percent. AI training workloads make it harder still, because thousands of GPUs surge and idle in near-synchrony, whipsawing demand on the power network.

Managing that requires capacitors at every stage. Film and electrolytic capacitors condition power at the front end, where the industry is moving from traditional 12-volt distribution toward 48-volt and even 800-volt architectures to cut losses. Polymer and tantalum capacitors provide bulk storage on boards. And dense arrays of small, high-capacitance MLCCs crowd as close to the processor as physics allows, on the board, on the underside of chip packages, and even embedded inside circuit boards to shorten the current path.

The component counts that result are unlike anything the industry has built before. TrendForce reports that a single NVIDIA GB200 compute board uses approximately 6,500 MLCCs, and that the next-generation Rubin architecture, with roughly double the power draw, will push per-board usage to around 12,000. Industry analyses compiled by the European Passive Components Institute estimate that an eight-GPU AI server requires 15,000 to 25,000 MLCCs versus about 2,500 in a standard server, and that a full GB200 NVL72 rack consumes roughly 440,000. Design churn amplifies the effect: TrendForce notes that during validation of AMD’s MI450 platform, aluminum electrolytic and tantalum capacitors on the bill of materials were replaced entirely with MLCCs, taking usage of one specific 47-microfarad part from 1,440 units per board to 10,544.

Wall Street has noticed. Goldman Sachs analysts have reportedly estimated that the AI-server MLCC market could grow from roughly $1.3 billion in fiscal 2025 to approximately $5.8 billion by fiscal 2030, a rate of expansion few component categories have ever posted. Critically, AI demand concentrates in exactly the specifications that are hardest to manufacture: the smallest cases, the highest capacitance, the widest temperature tolerance.

Who makes MLCCs?

Very few companies, and almost all of them in Japan, South Korea, and Taiwan. Murata Manufacturing leads the industry with a global share commonly cited near 32 percent, followed by Samsung Electro-Mechanics, TDK, Taiyo Yuden, and Yageo, which together account for roughly three-quarters of the market according to figures from the China Electronic Components Association cited in Korean media. Kyocera, Vishay, Walsin, and Chinese challengers such as Fenghua Advanced Technology round out the field. In the high-end grades that AI servers demand, concentration is even tighter: industry analyses estimate Murata and Samsung Electro-Mechanics together supply on the order of 84 percent of AI-grade MLCCs.

The barriers to entry are formidable and mostly invisible. Competing at the leading edge means formulating proprietary ceramic powders, casting dielectric layers thinner than a micron, stacking as many as a thousand of them with near-perfect registration, and firing the result at high temperature without cracks or shorts, at yields that make the economics work across billions of units. Even the leaders strain at the frontier: TrendForce reports that Murata only began mass production of the newest ultra-high-capacitance parts for AI at the end of 2025, with Samsung Electro-Mechanics following in March 2026, and that yields on these specifications remain an industry-wide challenge.

The 2026 supply squeeze

All of this collided in 2026. TrendForce data show book-to-bill ratios, the measure of orders received versus product shipped, at the three leading Japanese and Korean suppliers reaching their highest levels since the pandemic, with Murata at 1.30, Samsung Electro-Mechanics at 1.31, and Taiyo Yuden at 1.25 by late June. Murata’s orders-to-backlog ratio has surpassed the level recorded at the onset of the 2018 capacitor shortage, the most severe in the industry’s history. Lead times on key AI and automotive grades have stretched from a normal 8 to 10 weeks to 20 to 26 weeks, and reporting compiled by TrendForce indicates some high-end products are now quoted at 5 to 10 months, with tightness potentially extending into 2027.

Pricing has followed. Murata implemented increases reported at 15 to 35 percent on AI-server and automotive grades effective April 1, its first broad hike in roughly three years. Taiyo Yuden raised prices on select lines by 6 to 13 percent, and Korean media reports cited by TrendForce indicate Samsung Electro-Mechanics notified customers of a 30 percent increase on certain shipments beginning August 1. In China’s spot market, coverage from 36Kr describes broad gains of 15 to 20 percent since late February, with high-capacitance AI-grade parts up 50 to 60 percent. TrendForce has warned that as cloud providers ramp custom AI accelerators alongside GPU platforms, high-end specialty MLCCs face the risk of structural shortage in the second half of 2026. The pressure is spilling into neighboring technologies too, with reported price increases across tantalum and aluminum electrolytic capacitors and accelerating investment in silicon capacitors.

The bottom line

Passive components have historically been the afterthought of the electronics industry: purchased by the reel, priced by the fraction of a cent, and noticed only when they run out. The AI buildout is changing that. When a single rack of accelerated computing hardware contains nearly half a million capacitors, when the parts stabilizing every GPU’s power rail are made at scale by only a handful of companies, and when lead times are quoted in months rather than weeks, the humble capacitor stops being a rounding error and becomes a strategic input.

For anyone trying to understand AI infrastructure, the lesson is that compute is a system, and systems are built from parts. The processors get the headlines, but they only run because trillions of tiny ceramic chips hold their voltage steady, thousands of times per board, billions of times per data center. Knowing what a capacitor is, and why the MLCC in particular sits at the center of the current cycle, is quickly becoming table stakes for understanding the modern electronics economy.

Sources

• TrendForce, “CSP In-House ASIC Boom Drives MLCC Specification Concentration; Structural Shortages of High-End Specialty MLCCs May Emerge in 2H26,” June 2026
• TrendForce, “MLCC Market Polarizes in 1Q26 as Embodied AI Ignites High-End Demand,” February 2026
• TrendForce, “Strong AI Demand vs. Weak Consumer Segment Creates Pricing Divergence Among MLCC Suppliers,” April 2026
• TrendForce News, “MLCC Lead Times Diverge as High-End Products Stretch to 5-10 Months,” August 2026
• TrendForce Insights, “Powering Next-Gen AI: Capacitors’ Dual-Track Evolution (MLCCs and Silicon Capacitors),” June 2026
• TrendForce News, “Samsung Electro-Mechanics Reportedly Weighs Double-Digit MLCC Price Hike Amid AI Demand,” February 2026
• TrendForce News, “MLCC Turns From Electronic Rice Into a Gold-Like Rally,” citing 36Kr, June 2026
• European Passive Components Institute / passive-components.eu, “MLCCs in the Age of AI: Q2 2026 Market Tightness,” June 2026
• Astute Group, “MLCC Order Books Hit 2018 Levels as AI Servers Swallow High-End Capacitor Capacity,” August 2026
• Persistence Market Research, “Multilayer Ceramic Capacitor Market Size and Share,” 2026
• Bloomberg, “Murata Explores Raising Prices of Key AI Server Component,” February 2026 (as cited in industry coverage)
• Goldman Sachs and Morgan Stanley analyst estimates as reported in industry coverage, 2026