High-end MLCC prices rise by 15% to 35%: AI splits the "electronic rice" into two markets

An electronic component even smaller than a fingernail is seeing its price driven up by AI.

Recently, MLCC, known as the "rice of the electronics industry", has witnessed a very special market trend: for high-capacity MLCC used in AI servers, quotations from some original manufacturers have risen by 15% to 35%; some in-demand models were once speculated to several times their original prices in the spot market.

However, oddly enough, the ordinary consumer-grade MLCC on the other side has not risen along with it, and some products have even fallen back from their previous highs.

Both are capacitors. Why is there a shortage on one side and no significant supply crunch on the other?

The answer may be more significant than "AI demand is too strong":

AI is splitting a previously highly standardized electronic component market into two completely distinct markets.

In the future, when judging the prosperity of the electronic component industry, people may no longer simply ask "has the price of MLCC gone up", but instead ask:

Which type of MLCC is rising in price? Who is buying it? Where is it being used?

Why are AI servers inseparable from this component that costs only a few cents apiece?

MLCC, the full name of which is Multi-layer Ceramic Capacitor.

Ordinary consumers barely notice it, but from mobile phones, computers, home appliances to automobiles, communication equipment and servers, almost every electronic circuit cannot function without it.

One of its most important functions is to stabilize voltage, filter out noise, and store and release electric charge.

Therefore, the industry often refers to it as:

"The rice of the electronics industry".

This metaphor is very vivid.

A single MLCC is not noticeable, but hundreds, thousands or even more of them are often required in electronic devices.

In the past, smartphones were one of the most important application scenarios for MLCC, and a single mobile phone might require thousands of MLCC units. After the electrification level of automobiles increased, automobiles became a new source of demand.

Now, AI servers have arrived.

The problem is not only that AI servers need more MLCC, but more importantly, the MLCC they require is often not the most common type.

The power consumption of AI chips is getting higher and higher.

GPU, ASIC, high-speed memory and power systems will experience huge current changes in a very short period of time. If the voltage is not properly controlled, the stable operation of the chips will be affected.

Therefore, a large number of MLCC units need to be deployed around core chips such as GPU and ASIC to undertake the tasks of transient power supply and voltage stabilization.

This means that the higher the AI computing power, the more complex the power supply design, and the stronger the demand for high-capacity, high-reliability MLCC.

Current industry chain surveys show that after some AI workstations are upgraded, the MLCC consumption per unit has reached 2 to 3 times that of traditional equipment; new equipment such as high-speed optical modules are also increasing the usage of high-performance MLCC.

The construction of AI computing power has finally been transmitted to this seemingly insignificant small component.

This is one of the most noteworthy changes in today's electronic industry chain:

What AI has truly changed is not only the GPU market, but almost the entire value distribution of components inside servers.

Why aren't all MLCC prices rising?

If people only see that "the maximum price of MLCC has increased several times", they can easily draw a conclusion:

The entire industry is facing a supply shortage.

But the actual situation is far from that simple.

A recent report from CCTV Finance shows that this round of market trend presents a very obvious structural differentiation: high-capacity models used in AI servers remain tight, while ordinary consumer-grade products have fallen significantly after their previous rapid price increase.

Why is that?

Because although MLCC shares the same name, there are extremely complex specification differences inside.

As long as several parameters such as capacity, voltage, size, temperature characteristics, reliability and service life are different, the manufacturing difficulty and application scenarios of the products may be completely different.

The MLCC required by an ordinary household electronic product and the MLCC required by an AI server that needs to run at high load 7×24 hours are not in a simple substitution relationship.

The operating environment of AI servers is much harsher.

It has high power consumption, high temperature and large current, and once a failure occurs, the loss is far higher than that of ordinary consumer devices.

As a result, high-capacity, high-reliability, miniaturized MLCC has become more valuable.

According to public reports, for some high-capacity MLCC used in AI servers, the price increase by original manufacturers has reached 15% to 35%. As of August, the delivery cycle of some high-capacity, high-voltage products has reached 15 to 20 weeks.

This shows that the most accurate description of this round of market trend is not:

"Full-scale shortage of MLCC."

Instead, it should be:

"The supply of high-end MLCC cannot keep up with the demand changes brought by AI for the time being."

Such a slight difference in wording leads to completely different industrial logic.

The most powerful part of AI is that it drives up both "quantity" and "specification" at the same time

In the past, many industry chain analysts liked to calculate one indicator:

How many components does a single device need?

But in the era of AI servers, counting only the number of units may no longer be sufficient.

Because AI brings two types of changes.

The first one is of course the increase in quantity.

Higher device computing power, more GPUs, and more complex power modules naturally lead to an increase in the number of capacitors that need to be configured.

But the second change is even more important:

The specification of each single product is also being upgraded.

For example, occupying the same PCB area, high-capacity MLCC can provide stronger charge storage capacity. But to obtain higher capacity in an increasingly smaller volume, it is necessary to make the internal ceramic dielectric thinner and stack more layers, while ensuring the yield rate and long-term reliability.

This means that the problem of high-end MLCC shortage cannot be solved simply by "producing a few more units".

Equipment, materials, processes and yield rates all need to be improved accordingly.

Some industry studies estimate that high-capacity MLCC consumes significantly more production capacity than ordinary products, and as the number of product layers increases and the dielectric becomes thinner, the manufacturing difficulty also rises.

Therefore, the demand growth brought by AI is not the ordinary sense of demand expansion.

It is more like this:

In the past, the market needed 100 ordinary products;

Now it needs 150 units, and a considerable part of them must be upgraded to high-end products that are more difficult to manufacture.

In this case, the demand seems to have increased by 50%, but the actual occupied high-end production capacity may have increased even more.

This is also why even if the sales volume of AI servers does not increase tenfold, some high-end components will still quickly fall into a tight supply state.

"Price changes every hour", but the spot price should not be regarded as the industry-wide price

This round of MLCC market trend also has a very eye-catching phenomenon.

The spot prices of some models once rose to an extremely exaggerated level.

In June, some popular MLCC models in Huaqiangbei, Shenzhen even saw "price changes every day and every hour", and the price of a few models rose 7 to 8 times within a month. In extreme cases, the spot price reached 2 to 10 times the original price.

This figure is very staggering.

But if people therefore think that MLCC manufacturers have generally sold their products at ten times the original price, they are wrong.

There are at least several types of prices in the electronic component market.

The first type is the quotation from the original manufacturer.

The second type is the long-term contract price for large customers.

The third type is the price in the distribution and spot markets.

When a certain model suddenly falls into shortage, downstream customers are willing to buy goods at high prices in the market in order to meet the production schedule, and the spot price may soar instantly.

But this kind of price usually reflects the "urgency level", and does not equal the long-term transaction price of the entire industry.

A very typical example is that the price of a relatively common consumer MLCC on the market was about 5 yuan per 1000 units, once rose to around 30 yuan at the peak, and then fell back to about 10 yuan.

At the same time, the shortage of high-capacity MLCC for AI servers and high-end applications continues.

This once again proves that:

What really matters is not the two words "price increase", but who can maintain the high price after the price rise.

If the price of consumer-grade products rises rapidly and then falls back quickly, it indicates that the price change is more likely to be driven by channel restocking and market sentiment fluctuations.

If the price of high-end products keeps rising with a long delivery cycle and customers are still locking in production capacity in advance, it means that the demand structure has really changed.

A larger change: the electronics industry is bidding farewell to the "unified cycle of consumer electronics"

In the past, when analyzing the MLCC industry, a very important indicator was the sales of mobile phones, computers and home appliances.

Because these products have large sales volumes, they have a significant impact on the demand for electronic components.

When mobile phones sell well, the entire industry chain restocks;

When mobile phone sales decline, all upstream links go to destock together.

Therefore, many electronic components have obvious consumer electronics cycles.

But after the emergence of AI, this unified cycle is being broken.

On one hand, the growth of mature consumer electronics markets such as mobile phones and PCs is limited;

On the other hand, AI servers, data centers, high-speed optical modules and advanced automotive electronics are continuously increasing the usage of high-end components.

As a result, two completely different markets may even exist simultaneously within the same electronic component enterprise:

Ordinary products are still competing on cost;

High-end products are competing on production capacity.

Traditional products are still in the destocking phase;

AI-related products have started to lock in supplies half a year in advance.

This is also why when evaluating an electronic component enterprise today, looking only at the total production capacity is no longer sufficient.

The more important questions become:

How much of its production capacity can be used to manufacture high-end products?

Has it passed the customer certification?

What is the yield rate?

Has it actually entered the server supply chain, or is it still focused on ordinary consumer electronics?

These indicators are becoming more important than the simple figure of "annual output of hundreds of millions of units".

For Chinese manufacturers, the opportunity is not simply "filling the gap when others are out of stock"

Every time there is a shortage of electronic components, a familiar term will appear in the market:

Domestic substitution.

But the domestic substitution of high-end MLCC is not as simple as people imagine.

Because after electronic components enter high-reliability fields such as AI servers and automobiles, what customers value most is not just the price.

It is stability.

It is not difficult to produce a batch of qualified products. The difficulty lies in maintaining long-term consistency for billions or tens of billions of product units.

Behind this lies the ceramic powder, electrode materials, sintering process, lamination technology, equipment precision, and continuous and stable quality control.

In addition, entering the supply chains of large servers, automotive electronics and other fields also requires lengthy customer testing and certification.

Therefore, the real industrial opportunity is not simply:

Overseas manufacturers raise prices — domestic manufacturers immediately receive orders.

A more reasonable path may be:

After the supply of high-end MLCC becomes tight, downstream manufacturers are willing to provide more testing and introduction opportunities for the second and third suppliers for the sake of supply chain security.

This can create a real window of opportunity for local manufacturers.

In other words:

Being in a supply shortage does not mean competitiveness. Only by completing customer certification during the shortage can it be turned into real competitiveness.

The biggest risk is hidden in the two words "capacity expansion"

After prices rise and orders increase, the most natural next step is to expand production capacity.

But MLCC has gone through such cycles many times:

Supply shortage — price rise — capacity expansion — inventory increase — price fall.

Therefore, when seeing the current shortage of high-end MLCC, we cannot simply conclude that the entire industry will maintain high prosperity in the next few years.

What really needs to be observed are two speeds:

How fast is AI demand expanding?

And:

How fast can high-end production capacity be released?

If AI servers continue to grow rapidly, and products are continuously upgraded to higher power consumption and higher capacity, the demand side may continue to absorb the newly added production capacity.

But if a large number of manufacturers expand production at the same time, while the growth rate of AI equipment procurement slows down, the supply-demand relationship may change again.

In addition, there is another easily overlooked problem.

AI hardware itself is also constantly improving.

The power supply architecture, packaging method, chip design and capacitor configuration of servers may all change.

The number of MLCC units required by a server today does not mean that the same quantity and specification will still be needed five years later.

Therefore, for this industry, what is really worth tracking in the long run is never a single spot price surge.

Instead, it is:

The order visibility, customer certification, delivery cycle and actual capacity utilization rate of high-end products.

AI is re-pricing those inconspicuous components

In the past two years, the AI industry that has easily attracted the most attention is GPU.

Because it is expensive, scarce, and a single chip costs tens of thousands of dollars.

But after AI data centers are built on a large scale, people begin to discover a rule:

What determines whether an AI server can run stably is not only the GPU.

Power supply, heat dissipation, PCB, connectors, optical modules, storage, and those seemingly inconspicuous capacitors all need to be upgraded together.

As a result, the AI industry chain has entered the second stage.

In the first stage, people asked:

Who can produce AI chips?

In the second stage, more and more people are beginning to ask:

Who can equip increasingly powerful AI chips with a full set of high-reliability components?

This round of structural shortage of MLCC is one of the most intuitive examples of this change.

It does not mean that all MLCC will become more and more expensive, nor does it mean that the entire industry will bid farewell to the cycle from now on.

What is really happening is:

AI is redefining the value boundary between "ordinary electronic components" and "high-end electronic components".

In the past, all of them were called "the rice of the electronics industry".

Now, even the same "grain of rice" has completely different prices, production capacity and competition thresholds.

For China's electronics industry, what is really worth paying attention to is not how much the price of a single capacitor has risen.

Instead, after AI turns more and more ordinary components into high-end components, can we truly enter the market that is more difficult to manufacture, has stricter certification requirements and brings higher value.

This may be the biggest industrial signal left by this round of MLCC market trend.

This article is from the WeChat official account "BT Finance" (ID: btcjv1), written by BT Finance, and authorized for release by 36Kr.