Economic Observer reporter Wang Yajie

Hundreds of servers reached the final stage of installation and were stuck on the production line by several tantalum capacitors.

Zhou Zihang, a supply chain manager at a leading AI (artificial intelligence) server manufacturer in China, told Economic Observer reporters that a small tantalum capacitor "stuck" a batch of machines, which has happened several times this year.

The business manager in charge of tantalum capacitor business at a domestic listed company explained that in this computing power delivery chain, there is a shortage of chips and copper foil, which hinders the start of production. The transformer is missing, and the issue is with the power supply of the server after it is transported into the park. But if tantalum capacitors are missing, the stuck ones are the last step before the machine goes offline and leaves the factory. It is not so conspicuous, but it can make hundreds of complete machines stop on the production line.

Tantalum capacitors perform microsecond level power replenishment in servers. As soon as the AI server is turned on, the graphics processor that consumes kilowatts of power will instantly "draw" hundreds of amperes of current. Tantalum capacitors need to replenish the power and stabilize the voltage within microseconds. If it slows down by half a beat, the entire machine may shut down. So the usage of single unit tantalum capacitors has increased from a few dozen in traditional servers to two to three thousand in mainstream models.

The two hard constraints on the supply side are widening the gap in related computing power.

Both of these constraints are held overseas. One is resources. Most tantalum mines are located in Africa, and even the high-end tantalum powder used to make capacitors relies on imports. The second is production capacity, with over 70% of high-end production capacity held by relevant overseas giants.

The person in charge of the tantalum capacitor business mentioned above stated that for domestic computing power enterprises, the shortage of transformer supply is a problem of queuing for goods; The shortage of tantalum capacitors is an anxiety caused by the lifeline being held in someone else's hands.

Under the layers of transmission, the window for domestic substitution is being opened.

The person in charge of the tantalum capacitor business mentioned above told the Economic Observer reporter that the localization rate of tantalum capacitors for AI computing power is gradually increasing, and they will still be in a small batch delivery state in 2026. It is expected to climb significantly in 2027, and the real large-scale production will be from the end of 2027 to 2028.

A gap, spreading from the mine to the capacitor

Tantalum concentrate was the first to 'fall'.

More than 80% of the world's tantalum mines are concentrated in three African countries, and most of them are manually mined on a small scale, with extremely low supply elasticity. After the collapse and shutdown of the core mining area in the Democratic Republic of Congo in early 2026, about 15% of the world's tantalum raw materials disappeared instantly. The resumption of production in such mines has never relied on capital expenditures, but on the local safety and governance environment, which is precisely the most difficult factor to change in the short term.

Tantalum powder followed closely behind. The spot of midstream smelting is highly dependent on upstream mineral materials. As soon as the mining end tightens, the supply and bargaining space also tighten simultaneously. The technology and production capacity threshold of capacitor grade high specific capacitance tantalum powder is higher, and there are fewer alternative sources, further amplifying the shortage.

The most dead card is the finished capacitor at the downstream of the chain.

Global tantalum capacitor leader Jimei (under Guoju) has raised prices multiple times since June 2025. The latest round of price adjustments was sent in a letter in March 2026 and officially took effect on April 1, with a significant increase in prices for AI specific polymer tantalum capacitors. Panasonic and AVX have also followed suit, with a continued tight supply of high-end models.

The first to feel the pain of this shortage is the whole machine factory.

Zhou Zihang said, "The regular delivery time of overseas tantalum capacitor manufacturers is being extended, and the price has significantly increased compared to the same period last year

In order to grab the goods, Zhou Zihang's company has started to lock in long-term framework orders, predicting the demand for the next 6 to 12 months in advance, and signing a framework agreement with the original tantalum capacitor factory to lock in production capacity. But even if long-term orders are locked in, the production capacity allocation of original factories (such as Jimei, AVX, Panasonic and other overseas brand original factories) is still tight, and it is still difficult to completely avoid the risk of supply interruption.

The path of safety stock is also stuck by specifications.

Zhou Zihang added that tantalum capacitors have a wide range of specifications, and if the voltage, capacitance value, and equivalent series resistance parameters are almost the same, they cannot be mixed. It is impossible to stack all specifications of products into inventory, and the factory's funds and storage cannot withstand it. Only a small inventory buffer can be made for high-frequency essential models.

What is even more headache inducing for the whole machine factory is that this material is "small" and inconspicuous, but the risk is not small.

According to Zhou Zihang's analysis, from the cost proportion of the Bill of Materials (BOM), the complete set of tantalum capacitors in an AI server usually accounts for less than 1%, which is almost negligible compared to GPUs and CPUs (central processing units). However, it is a key component in the power supply chain, and a single failure can cause a single board failure, making the entire machine unable to power on and operate.

The server OEM is also re scheduling production.

Zhao Mingxuan, the supply chain project leader of a server OEM factory, also told Economic Observer reporters that the delivery time of tantalum capacitors has been extended to 40 to 52 weeks, which exceeds the stocking cycle of conventional server projects. Semi finished machines can only be stacked in online side warehouses, and when the shortage of materials is filled, they can return to the production line. The disorderly order of work orders directly reduces the utilization rate of the production line.

Zheng Guodong, the business leader of a leading tantalum powder enterprise, told Economic Observer that the prices of tantalum powder and tantalum wire have risen synchronously with the raw material end, but the shipment volume has not experienced explosive growth. The selling price of the product has risen with the tantalum concentrate market, but it has not been a one-way surge. It has also experienced price fluctuations caused by downstream observation and bargaining games. The fluctuation of mining prices on the cost side will greatly squeeze the processing profits of the enterprise, and the raw material side is not as profitable as the outside world imagines.

In Zheng Guodong's view, the increase brought by the computing power track is tangible. At present, the proportion of orders directly supplied to tantalum capacitor customers related to computing power is not particularly high, but the growth rate is very prominent, with a significant year-on-year increase higher than traditional sectors such as consumer electronics and military aerospace.

Need to collide with rigid supply

In January 2026, the Lubaya mining area in the eastern Democratic Republic of the Congo (DRC) collapsed due to rainstorm, and this mining area, which accounts for about 15% of the global tantalum supply, will stop production. The resumption of production involves safety rectification and large-scale infrastructure construction, and it will be difficult to return to the mining state in 6 to 12 months.

The mine has stopped, but demand has not stopped.

Zheng Guodong told the Economic Observer reporter that the usage of single tantalum capacitors in AI high-power servers has doubled, which has been transmitted to the upstream and has significantly driven the demand for high specific capacity tantalum powder.

There is actually not much that the supply side can do. After the shutdown of the Rubaya mining area, the spot price of tantalum concentrate quickly rose, and the market circulation of spot goods tightened, resulting in a phenomenon of grabbing goods.

Zheng Guodong introduced that his company does not rely solely on the source of goods from the Democratic Republic of Congo. It has built a global procurement system based on the group's resource layout, and has allocated sources through multiple channels. There has been no interruption in production. At the same time, it has signed medium - to long-term procurement contracts with overseas mining enterprises for iron tantalum niobium alloy sources as a stable raw material supplement, and has increased its recycling efforts. The tantalum waste inside the factory is recycled in a closed loop, and industrial end tantalum scraps and scrapped capacitor dismantling materials are purchased externally.

But overseas mines and recycled materials can only meet a portion of the demand.

A representative from a state-owned metal mine told the Economic Observer that tantalum is a rare and scattered metal that is mostly associated with other mines. It is difficult to quickly build new mines and expand production just because tantalum prices have increased. African mines are mainly engaged in manual and small-scale mining, and there is great uncertainty in resuming production due to multiple factors such as safety, geography, and policies; The release cycle of newly added primary minerals generally takes several years.

Beyond the mining sector, downstream production capacity is to some extent locked in. Jimei AVX、 The expansion focus of the four overseas giants, Weishi and Panasonic, is on automotive grade and industrial grade tantalum capacitors, with limited increase in AI grade production capacity. According to feedback from relevant supply chain personnel, major overseas cloud providers such as Microsoft, Google, Amazon, Meta, and other large-scale data center operators, as the ultimate demanders of AI servers, often bypass the whole machine factory and directly sign long-term framework orders with capacitor manufacturers to ensure their own computing power expansion, locking in more than 70% of the high-end production capacity mentioned above.

A tantalum capacitor production line takes 18 to 24 months from construction to mass production, and there is basically no new supply before 2028.

Where is the mismatch reflected?

The above-mentioned state-owned enterprise personnel in the metal mining industry explained that firstly, demand is fragmented. Different AI server models have different indicators for tantalum powder particles, purity, and porosity. Downstream capacitor enterprises require many customized specifications, but upstream smelting enterprises are more willing to produce standardized products in large quantities. Small batch customized orders have high costs, large R&D investment, and weak willingness to actively accept orders; Secondly, there is a barrier to information communication between the whole machine, capacitor factory, and tantalum powder enterprise. It is difficult to fully transmit the requirements of terminal operating conditions to the upstream, and there are often situations where sample parameters meet standards and batch production consistency does not meet standards; The third issue is that the verification chain is too long. If the upstream material changes one process parameter, the downstream will have to complete one round of reliability verification, which often takes more than a year. The cost of iterative trial and error is very high.

It is precisely because the supply side cannot make up for it temporarily, and the mismatch cannot be resolved temporarily, that it is difficult to reverse the tantalum price increase in the short term.

According to the judgment of the above-mentioned central enterprise personnel in the metal mining industry, the recent price increase of tantalum as a computing power metal is not entirely short-term speculation, but a structural market caused by the combination of rising demand and supply disturbances. He predicted, "The possibility of a simple short-term decline is unlikely, but it will not continue to rise unilaterally

Domestic tantalum capacitors break through

A domestic tantalum capacitor enterprise with high reliability component research and development accumulation is spreading mature reliability technology to the computing power track.

The business personnel of this enterprise told the Economic Observer reporter that its tantalum capacitor products have achieved small-scale supply to domestic AI servers. Capacitors in computing power scenarios require long-term 7 × 24 hour high-temperature operation, withstand repeated large ripple currents, and face significant pressure for batch consistency.

He Junfeng works at another domestic tantalum capacitor factory that has already delivered small quantities. He told the Economic Observer reporter that it took the company 14 to 18 months from sample delivery to achieving small batch delivery. In the early stage, the company needs to complete laboratory parameter comparison, board level testing, high and low temperature cycling, long-term full load aging, and then small batch trial production and whole machine joint debugging. The customer needs to install the capacitor on a real AI whole machine and run 7 × 24-hour pressure testing for several months.

Consistency is the hardest hurdle for domestic materials.

He Junfeng said that a single sample of tantalum capacitor can be produced very well, but if produced in batches of tens of thousands of pieces, the parameter dispersion between different batches will be higher than that of overseas large factories. What customers are most concerned about is not static nominal parameters, but stability under long-term operating conditions, leakage current drift under high-temperature loads, and equivalent series resistance rise after thousands of hours of aging.

He Junfeng said, "If all single samples of tantalum capacitors pass the inspection, and there are individual abnormal samples found in small batches and multiple batches of verification, the customer will directly suspend the import (i.e. suspend the certification process of this model entering the formal material list)

Powder is a hurdle upstream.

Zheng Guodong added to the Economic Observer reporter that the working conditions of AI servers are demanding, and the requirements for tantalum powder are much higher than those for ordinary consumer electronics. They not only pursue high specific capacitance (specific capacitance refers to the amount of capacitance that can be provided per unit volume or unit mass, which means that devices can be made smaller under the same capacitance value and is a key indicator of high-density integration of AI servers), but also value the high consistency of performance between batches. The impurity content, particle size distribution, and porosity should be highly stable, and trace impurity fluctuations can cause abnormal leakage current of tantalum capacitors, directly leading to server board failures.

He Junfeng also pointed the ceiling to the powder. He stated that the domestic supply chain for mid to low end tantalum powder has matured, but the ultra-high specific capacity tantalum powder suitable for AI server polymer tantalum capacitors has extremely high requirements for impurity control and particle batch consistency. Domestic powders are still iterating and climbing, and the company's high-end models still need to be matched with some imported tantalum powder. Even if the equipment, process, and customer certification are all in place, if stable high-end tantalum powder cannot be obtained, the production capacity of computing power tantalum capacitors cannot be released on a large scale.

The support of policies falls on both the research and verification ends.

A person close to the Ministry of Industry and Information Technology told the Economic Observer that high reliability tantalum capacitors are mainly developed through industrial infrastructure reconstruction projects, focusing on key materials such as upstream ultra-high specific capacitance tantalum powder and tantalum wire, as well as polymer tantalum capacitor body processes and reliability verification systems. At the same time, efforts are being made to improve the domestic component verification public platform and reduce the testing costs of domestic material sampling and iteration.

But the increase in domestic materials cannot be accelerated.

He Junfeng believes that there is no shortcut to the certification of AI server components. From sample delivery, multiple rounds of reliability testing, small batch trial production, to entering the formal material list, it generally takes 1 to 2 years. Even if the production capacity is built and customer certification is not completed, the production line can only be idle. Domestic substitution is gradual and difficult to achieve overnight.

(Author Wang Yajie)