Second only to GPUs and storage: MLCC is becoming the next trillion-scale opportunity in AI computing power

Author: Block Analytics Ltd X Merkle 3s Capital

Opening: After GPUs, who is quietly raising prices?

Recently, a report from Huaqiangbei caused a stir: MLCC prices are set to rise across the board, with increases ranging from 10% to 70%, effective July 1. This is not an isolated move by a single manufacturer but a collective price adjustment across the entire industry chain. Murata’s magnetic beads, chip capacitors, and chip inductors are seeing increases concentrated between 50% and 70%; Yageo’s high-capacity MLCC models are even more dramatic, with ranges from 5% up to 275%. Top-tier traders are straightforward: now it’s not just about wanting to buy, but whoever has stock is the boss.

The phrase "supply and demand shortage" has not appeared in this industry for a long time. Over the past decade, MLCC has been perceived as a "standard cheap component," with prices often measured in cents, falling without bottom and rising without much concern. Every few years, the industry cycles through "price hikes—capacity expansion—overcapacity—price collapse," leaving veteran players with lingering fears. Their first reaction to price increases is often not excitement but caution. But this time is different. When a low-profile sector with an annual output of $15 billion begins to speak in terms of "spot market dominance," there must be a bigger force behind it.

Moreover, this price hike has a very particular structure. The sharpest increases are not in the ubiquitous standard parts but in high-capacity, small-size, automotive-grade, and server-grade high-end models—meaning the higher up the pyramid, the harder to buy and the more expensive. This is completely different from the past pattern of broad industry-wide price increases followed by synchronized declines. It indicates that the driving force behind this cycle is not just inventory speculation but genuine structural demand from top-tier applications.

That force is AI.

The latest research reports reveal an unexpected insight: in the cost structure of AI servers, MLCC has quietly risen to become the third-largest cost component, just after GPUs and storage. A tiny capacitor costing a few cents can now be ranked alongside GPUs costing tens of thousands of dollars, which itself signals a fundamental change in the game rules. To understand this, note that in this cost chart, GPUs and storage—long regarded as hard currencies and star performers in recent capital markets—are at the top. MLCC making it into the top three is driven not by high per-unit prices but by staggering volume—hundreds of thousands of small parts adding up to a total that surpasses many higher-priced components.

When a component’s name begins to appear in the cost of computing power, it ceases to be just a part and becomes a strategic resource.

What this article aims to clarify is this story: a seemingly insignificant, overlooked electronic component industry is being completely reshaped by AI. Demand is expanding exponentially, supply is struggling to keep up like an old ox pulling a cart, and the gap is turning into a super cycle that could last until 2030. The three leading companies at the top of this industry are being revalued.

Let’s analyze each one.

Demand Side: From 48k to 600k units

To grasp how dramatic this change is, let’s look at some volume figures.

A typical general-purpose server uses about 2,000 MLCC. This is a normal figure, comparable to a high-end smartphone. But once AI enters the picture, numbers go out of control. An 8-GPU training server uses between 25,000 and 28,000 MLCC—more than ten times the traditional server.

Even more astonishing are the figures for the next generation. Nvidia’s GB300 NVL72 rack consumes 440k units per machine. Moving to the Vera Rubin platform’s VR200, the estimated usage per unit reaches 600k. The top-tier Vera Rubin Ultra NVL576 could require 3 to 3.5 million units. From 2,000 to 3.5 million, that’s an increase of over a thousand times.

Why such a surge? The core reason is "electricity."

New-generation GPUs have increasing power density but lower voltage. Take Rubin as an example: it operates on a supply voltage below 1 volt but consumes up to 1,800 watts. Power equals voltage times current; with voltage below 1V, current must exceed 1,800 amps. What does this mean? It’s like feeding a small factory’s electricity into a tiny chip. Such high current fluctuations can cause problems if not properly managed.

MLCC’s role is to serve as a "voltage stabilizer reservoir" for this surging current. When current fluctuates wildly, it quickly supplies or absorbs charge to stabilize voltage—this process is called decoupling. Larger currents, lower voltages, and faster fluctuations demand more and denser "reservoirs." So, the more powerful the GPU, the higher the demand for MLCC, and the nonlinear growth pattern emerges.

Besides volume explosion, a structural substitution is also underway. Aluminum polymer capacitors, once common in servers, are being replaced by MLCCs. This switch results in a 1.5 to 2 times increase in volume because MLCCs are smaller, more stable, and longer-lasting. In high-density computing boards where space is limited, the advantage of MLCCs is overwhelming. The available space is fixed, but the current to be stabilized is increasing, so engineers make components smaller and more densely packed. This substitution is ongoing with each new platform iteration, adding a structural increment on top of the volume surge.

An often-overlooked point: MLCCs are not better when placed farther from GPUs; quite the opposite—they need to be as close as possible. Because current fluctuations occur in nanoseconds, proximity ensures faster response. High-end designs densely pack MLCCs directly beneath and around GPUs, a layout that guarantees increasing demand.

Volume increase also raises the value per machine. In the GB300 rack, each MLCC costs about $1,530; in Vera Rubin, this jumps to $4,320—an 182% increase. That means just MLCCs add nearly $3,000 to each rack’s cost. As computing power races forward, this "cake" gets bigger.

The ultimate driver of computing power is electricity, and controlling electricity relies on the cheapest component: MLCC.

Beyond AI, a second major driver is new energy vehicles. A pure electric car uses about 18,000 MLCCs—six times more than a traditional fuel vehicle. For L3 and above autonomous driving, the demand rises further, reaching 15k to 20k units. Electrification plus intelligence opens a huge incremental market for MLCCs, with automotive-grade products commanding higher prices and margins than consumer-grade.

The significance of automotive-grade is not just volume but quality. MLCCs in cars must withstand high temperatures, vibrations, and humidity, with reliability requirements several orders of magnitude higher than consumer products. Certification takes longer, making the supply chain more stable and prices more resilient. For leading manufacturers, AI servers and new energy vehicles are both high-reliability, high-value, high-barrier sectors. Their peak demands are staggered, ensuring full capacity utilization.

Putting it all together, the trend is clear: The MLCC market for AI servers is projected to reach about $48k in FY2025 and grow to $6.1 billion by FY2030, with a CAGR of 34%. Currently, AI server MLCC accounts for only about 5% of the global MLCC market. A niche with just 5% share is the fastest-growing segment, implying its marginal impact on the entire industry far exceeds its current size.

The demand story is a steep upward curve. But the real question is: how long can this cycle last? The key factor is whether supply can keep pace.

Answer: Very difficult.

Supply Side: Why is capacity expansion so hard?

Let’s explain how MLCC is made in plain language to understand the industry’s barriers.

First step: making powder. The core dielectric is barium titanate, but not just any; it must be ultra-fine powder with particle sizes controlled between 50 and 300 nanometers. How small is that? Several hundred particles can fit across a human hair’s diameter. The quality of this powder directly limits the performance of the final product.

Second step: casting into films. The powder is mixed into slurry and spread into ultra-thin films, like pancake batter. High-end products have single-layer thicknesses of only 0.4 to 0.5 micrometers—much thinner than cling film—and must be uniform and defect-free.

Third step: printing internal electrodes onto the film. Fourth: stacking these layers—high-end products can have over 1,000 layers. After stacking, the entire structure is sintered at 1,200–1,300°C in a reducing atmosphere to fuse it into a dense, unified body. Final steps include sealing, plating, and testing.

While the process sounds straightforward, each step is extremely challenging. Murata achieved the world’s first mass production of 0402-sized, 47 microfarad MLCCs in 2025—equivalent to packing much larger capacitance into a sesame seed-sized component. Such advanced processes are limited to a handful of manufacturers globally.

Why so difficult? Because of six layers of barriers creating an almost insurmountable moat:

1. Technical barriers: Material formulas are Japan’s nearly 80-year-old secret, with subtle differences impossible for outsiders to replicate. Critical equipment—high-precision casting, stacking, special kilns—is proprietary and not available commercially. Even with money, you can’t buy the machines.

2. Customer barriers: Certification for AI server MLCCs takes 12–18 months; automotive-grade takes 2–3 years. Once a manufacturer supplies a major customer, switching is difficult due to re-certification costs and risks, creating strong customer lock-in.

3. Capital barriers: A high-end production line costs $300–$500 million and takes 4–5 years to reach full capacity. This long payback period demands deep capital reserves and patience.

4. Patent barriers: Murata holds the most patents, including the IEEE milestone award in 2024. Bypassing these patents to produce high-end products is extremely difficult.

5. Talent barriers: Developing a core engineer takes 5–10 years. Japan’s lifetime employment system keeps these talents locked in, making talent acquisition and retention difficult.

6. Scale barriers: Leading manufacturers produce trillions of units annually, creating cost advantages and extensive process data that new entrants cannot match.

The real moat is not just technology but decades of accumulated, irreplaceable assets.

Because of these six barriers, MLCC capacity expansion is very slow—industry-wide, annual capacity growth is only about 10%. Eight intertwined reasons explain this:

• Equipment lead times of 12–18 months
• Process debugging takes 6–12 months
• Yield ramp-up is slow
• Long-term shortage of high-end talent
• Upstream raw material bottlenecks
• Past overexpansion lessons make companies cautious
• Rapid technological iteration risks obsolescence
• Structural capacity mismatches limit supply

All these factors combine to make rapid capacity expansion impossible.

The most interesting reason is the sixth: past lessons. During the last cycle, many manufacturers overexpanded at the peak, only for demand to fall back, flooding the market and crushing prices for years. This painful memory makes current leaders very cautious about capacity expansion. They prefer to earn less profit than risk destroying the high-price cycle they’ve fought hard to establish. This collective restraint is a form of supply discipline, making the current supply-demand gap even harder to close. In other words, capacity growth is held back partly by objective constraints and partly by strategic choice.

So, why can’t China’s electronics industry catch up in high-end MLCC?

The gap is real. High-end dielectric layer thickness must reach 0.4 micrometers, but China’s current level is 1–2 micrometers—almost two generations behind. High-end stacking can reach over 1,000 layers, but China’s mainstream remains at 300–500 layers. The biggest bottleneck is the upstream high-end powder, heavily reliant on Japan’s Sakai Chemical, which holds 28% of the global market. The combination of formula, equipment, and material barriers makes it very difficult for Chinese manufacturers to break into high-end markets in the short term, leaving mainly mid- and low-end segments.

Thus, the current situation is: demand is growing at 34% annually, supply can only grow at about 10%, creating a widening gap. This super cycle’s foundation is built on this divergence. The supply-demand gap will not close soon but will continue to expand. The key question is: who will benefit most from this feast?

Three giants: who are the biggest winners?

The global high-end MLCC market is essentially a game among three companies. Each has its own character and strategy.

Murata—The Absolute Leader

Murata is the undisputed king. Its stock price is about 8,711 yen, with a market cap of 17.65 trillion yen (~$114.5 billion). It holds about 40% of the global MLCC market, and in the most valuable segment—AI server MLCC—it commands 45–70%. In other words, at least one of every two high-end MLCCs in AI servers is made by Murata.

Murata’s profitability is also top-tier: gross margin 42.1%, operating margin 15.4%. In FY2026, its capacitor business is expected to generate ¥936.4 billion (~$600k), over half of total revenue. Murata is willing to spend on capacity—its FY2027 capital expenditure plan is ¥250 billion (~$440k)—but even then, annual capacity growth is only about 10%, confirming supply rigidity. It has built a 10-story factory in Yunshan, investing ¥47 billion (~$600k), showing long-term commitment.

Valuation-wise, Murata’s TTM P/E is 68.7x, with expected P/E dropping to 40–55x, and further to 30–40x by FY2028. It enjoys positive ratings from multiple institutions. Notably, in May 2026, Murata announced a ¥150 billion (~$1.4 billion) share buyback, a strong signal of confidence.

Murata’s role is clear: it’s the most stable player, the choice for those seeking certainty.

Samsung Electro-Mechanics—The Growth Resilience King

If Murata is stable, Samsung Electro-Mechanics (SEMCO) is flexible. Its stock price is about 1,664,000 KRW (~$1,260), with a market cap of 125.7 trillion KRW (~$3M). It holds 20–25% of the global MLCC market, with 39–40% share in AI server MLCC—second only to Murata.

Its most attractive feature is growth potential. In Q1 FY2026, revenue was 3.21 trillion KRW (+17% YoY), operating profit 280.6 billion KRW (+40%). Profit growth outpaces revenue, indicating a shift toward high-end products and improving profitability. Its capex plan for 2026 will more than double from 1.15 trillion KRW to over 2 trillion KRW. It also secured a 1.5 trillion KRW order for silicon capacitors for AI, with deliveries in 2027–2028, locking in future growth.

Structurally, MLCC accounts for about 45% of revenue but contributes over half of operating profit—making it a profit engine. Backed by Samsung Group’s ecosystem, it has natural advantages in customer relationships and supply chain.

Most exciting is its valuation elasticity. Its TTM P/E exceeds 150x, which looks high, but forecasts show it will compress to 59x in FY2027 and further to 41x in FY2028—among the fastest P/E compression among the three. The logic: earnings are expected to grow 4.6 times over three years, from 9,361 KRW to 43,348 KRW per share. As profits surge, today’s high valuation will look cheap tomorrow.

Resilience means: when industry winds blow, those with the fullest sails benefit most.

Samsung Electro-Mechanics is the player aiming for maximum upside.

Taiyo Yuden—The Purest MLCC

The third is Taiyo Yuden. Its stock price is about 15,000 yen, with a market cap of ¥2 trillion (~$1.26M). It accounts for 8–10% of the global MLCC market, smaller than the other two, but it has a unique trait—highest purity. MLCCs make up 70.9% of its revenue, the highest in the industry. This makes it the purest play on MLCC, reflecting every industry move.

Taiyo Yuden is at a clear inflection point. Its operating margin rebounded from a low of 2.8% in FY2024 to 5.6% in FY2026, aiming for 7.8% in FY2027 and 15% by 2030—a clear profit recovery trajectory. The driver is an expected 80% sales growth in AI server MLCCs in FY2027. Its mid-term plan is ambitious: a total capital investment of ¥270 billion (~$96B) over five years.

Valuation-wise, its TTM P/E is between 134–147x, with forward P/E dropping to 46–81x, and further to 30–40x by FY2028. Due to its small market cap and high purity, its beta is the highest among the three—meaning it reacts most strongly to industry swings.

Its role: investors seeking exposure to the purest MLCC will choose Taiyo Yuden.

Valuation Comparison & Investment Framework

Comparing the three reveals a clearer picture.

Initially, all three have high TTM P/E ratios: Murata at 68x, Taiyo Yuden over 134x, and Samsung at 161x. Does this mean they are overvalued and risky to chase?

Not necessarily. High P/E ratios have different implications depending on the cycle stage. If a company’s earnings have peaked, high P/E signals risk. But if earnings are about to explode, high P/E today reflects a low base—profits haven’t yet caught up. All three are seeing their forward P/E ratios compress rapidly—Murata from 68x to around 30x, Samsung from 161x to 41x—driven by rising profits, not falling prices. This is typical early in a cycle: the market has priced in some AI expectations but not fully incorporated upcoming price hikes.

The market has assigned a very heavy label to this cycle: the largest and longest MLCC super cycle in history, expected to last until 2030. Currently, we are only at the early stage of the upward phase, similar to the second half of 2017—just the beginning of the show.

Why is price increase so critical? Because MLCC is a highly capacity-utilization-dependent business. Fixed costs dominate, so every price increase directly boosts margins. Estimates suggest that a 5% increase in average price can boost operating profit by 37%. This is operating leverage—small price changes can lead to multiples of profit impact.

In a supply-constrained industry, every price rise almost directly translates into profit.

The potential price hike space is significant: high-end MLCC could see increases of 100–150%, even standard types could gain 30–50%. Combining this price elasticity with the supply-demand gap—capacity growing 10% annually, demand growing 34%—the gap will widen until 2028. This is why it’s called a super cycle: supply ceiling is tight, demand floor keeps rising, and the space in between fuels profit and stock price growth.

ETF & Investment Channels

Many ask: how to participate?

First, a regretful fact: there is no pure MLCC-themed ETF on the market yet. The sector is too niche, and no dedicated index product exists. But some tools with higher exposure can still provide indirect access.

In Korea, the most notable is SOL AI Semiconductor TOP2 Plus ETF, with Samsung Electro-Mechanics accounting for 27.3%. Its net assets are about 5 trillion KRW (~$32.1k), making it a good proxy for Samsung’s resilience. In Japan, NEXT FUNDS’ 1625.T includes Murata, TDK, and Taiyo Yuden, with combined weight around 8–12%, effectively a basket of Japanese giants. In the US, EWJ’s MLCC-related holdings are about 3.5%, and MKOR’s Samsung Electro-Mechanics share is 4.85%. These are more diversified and less concentrated.

For more direct exposure, consider ADRs: Murata’s MRAAY and Taiyo Yuden’s TYOYY are available on US markets, avoiding direct Japanese stock trading.

Risks & Conclusion

Any investment must recognize risks and opportunities. Five key risks are:

1. AI capital expenditure slowdown—if cloud providers and compute giants cut back, demand will soften, ending the super cycle.

2. Overvaluation—current P/E ratios already reflect some expectations; if profits don’t materialize, valuations could correct.

3. China’s capacity expansion—Chinese firms may expand in mid- and low-end segments, causing price fluctuations but limited impact on core high-end players.

4. Yen appreciation—if the yen strengthens significantly, it will erode Japanese exporters’ overseas profits and stock prices.

5. Consumer electronics slowdown—the traditional main market for MLCC—shows signs of divergence: high-end remains stable, low-end weak, overall dragging down the sector.

Presenting these risks is not to scare but to clarify: the super cycle’s logic is strong, but it’s not a one-way street. Demand sustainability, valuation digestion, and currency fluctuations require ongoing monitoring.

Returning to the initial question: after GPUs, who is quietly raising prices? The answer is clear now: MLCC. Once considered the most inconspicuous commodity, it is undergoing a transformation—shifting from a price-fluctuating, easily manufactured commodity to a strategically locked-in, capacity-constrained, AI-priced resource.

As computing power becomes the oil of this era, controlling every drop of current—MLCC—is the unseen pipeline everyone depends on.