High-capacitance MLCCs for AI servers: Chinese supplier progress and global benchmarks
AI-server demand is creating two different openings for Chinese MLCC manufacturers. Some can gain orders for general-purpose components as international suppliers allocate more capacity to AI products. Others are developing high-capacitance parts for demanding power-decoupling positions close to computing chips. Both opportunities can increase sales, but they require different qualifications and have different cost and profit implications.
High-capacitance multilayer ceramic capacitors provide local transient current as chip loads change. Limited board and package space makes capacitance density, usable capacitance under DC bias and long-term reliability important. For international engineering and sourcing teams, the useful distinction is between an available specification, repeatable production and a part qualified for a particular application. Recent Chinese supplier disclosures show progress across these stages, without establishing that all suppliers are ready for the same AI-server designs.
October market conditions diverged by product class
TrendForce's 29 October 2026 research reported an approximately 5% decline in spot prices for consumer low- and mid-capacitance MLCCs in early October. Price-increase momentum for high-end specifications had weakened, while AI-grade supply remained tight. The firm expected high-end shortages potentially to continue into the first half of 2027. That last statement is a forecast, not an observed extension of lead times.
Figure 1. Findings from TrendForce's 29 October 2026 research. The spot-price observation, supply assessment and forecast are shown separately; they do not describe every specification or customer transaction.
Samsung Electro-Mechanics' third-quarter results, released on 28 October, provide company-level evidence of demand. Component-division sales rose 36% year on year. The company planned to expand small, ultra-high-capacitance AI product supply in the fourth quarter and address demand for long-term supply agreements. The growth rate covers the whole division, not AI high-capacitance MLCCs alone.
For Chinese suppliers, redirected consumer-electronics and general-server orders can improve factory utilisation. Direct entry into AI-chip decoupling positions requires qualification of the relevant part under its actual operating conditions. TrendForce's September discussion of qualification activity involving Chaozhou Three-Circle and VIIYONG retained consumer-electronics and general-server application scopes; it does not prove volume supply to core AI boards.
The distinction affects investment returns. General-purpose orders remain exposed to end-market demand and competing supply. Higher-end AI products may improve the sales mix, but entail materials development, screening and customer-validation costs before those benefits are realised.
Chinese capacity and customer qualification are advancing at different speeds
Fenghua Advanced Technology's October disclosure addressed performance optimisation. It described development of an 0805-X6S-100 μF product and custom improvements in effective capacitance and high-temperature reliability. Its DC-ageing material showed effective-capacitance improvements of 8% and 25% at 30% and 60% of rated voltage respectively.
Figure 2. Original Fenghua test material published in October 2026. Reading guide: Ur denotes rated voltage; the horizontal axis is logarithmic time. The reported improvements are 8% at 30%Ur and 25% at 60%Ur. These are before-and-after results for the company's own process. Full test conditions were not disclosed, so the chart cannot rank suppliers.
This addresses usable capacitance rather than the nameplate value alone. A nominal 100 μF part may deliver less capacitance under DC bias, with temperature and ageing also affecting performance. Consistency in dielectric materials, internal electrodes, layering and firing influences both electrical behaviour and manufacturing yield.
Chaozhou Three-Circle's first-half 2026 report described higher MLCC sales volume and revenue and price recovery in some specifications. Cumulative investment in its high-capacitance MLCC expansion project was approximately RMB 1.861 billion. The date for reaching the intended usable state was extended to 31 May 2027. Investment measures project scale; its return still depends on qualified output, customer adoption and sustained orders.
VIIYONG's May AI-computing portfolio listed 0805, 100 μF, 2.5 V, X6S and 1206, 220 μF, 2.5 V, X6S products. Its June material reported stable volume production of principal high-capacitance models. Here, 0805 and 1206 are imperial case codes. These disclosures demonstrate specification and manufacturing progress, while a replacement assessment still needs thickness, effective-capacitance and reliability data.
EYANG's listed parent reported that several high-capacitance and high-reliability specifications had passed leading-customer qualification and entered volume supply. It also reported research and delivery progress for AI servers, computing and memory chips. The report did not map each customer to a part number and server platform, limiting how specifically adoption can be described.
Sunway Communication's half-year report similarly distinguished stages. Several high-capacitance MLCCs were shipping in batches to leading Chinese customers, and selected high-end parts were in stable volume production. Large-size, ultra-high-capacitance products had completed core-technology validation. The latter should not be counted as established production supply.
Other manufacturers are expanding the supply base. YUNZHONG is pursuing MLCCs through its Chizhou operation; its half-year report associated revenue growth with capacity release and demand while describing continued investment in high-capacitance and high-reliability development. Lihexing's MLCC activities include Lihexing Electronics and Lirong. Its report discussed high-capacitance research, commissioning of an internally developed stacking machine and expansion preparations. Further order and revenue evidence is needed to establish the scale of their core AI high-capacitance businesses.
Jiangsu Xinsheng Microelectronics is a Taiwan-invested manufacturer operating in mainland China. Its October product material continued to emphasise high-capacitance, high-voltage and automotive development. A September Huai'an project report described a second phase targeting AI servers and new-energy vehicles, with production expected in the first quarter of 2027. Development priorities and a construction timetable do not mean the planned output is already available.
High-reliability and RF suppliers serve distinct circuit requirements
The broader Chinese ceramic-capacitor base also includes specialists in high reliability, RF and microwave applications. Their capabilities can be relevant to AI infrastructure, but should be mapped to the actual circuit function.
| Supplier or group | Recent disclosed position | Relevance to high-capacitance AI decoupling |
|---|---|---|
| Torch Electron | High-reliability and commercial ceramic capacitors; a rising commercial share of self-produced component revenue | The disclosed share includes multiple component categories, not only AI high-capacitance MLCCs |
| Hongyuan Electronics | High-reliability ceramics and commercial high-end RF/microwave MLCCs and single-layer capacitors | Optical-communication and packaging activities need separate assessment from high-capacitance chip-power decoupling |
| Zhenhua Technology | High-reliability components; an ultra-miniature MLCC dielectric-material project in its half-year report | Materials progress does not establish a volume of AI high-capacitance shipments |
| Hongming Electronics, including Chengdu Hongke | High-reliability MLCCs and expansion into commercial applications | Parent and subsidiary are considered together; existing reliability capability does not itself establish AI-platform adoption |
| Hongda Electronics | Multilayer ceramics, single-layer ceramics and ceramic thin-film circuits | AI-related business growth cannot all be attributed to multilayer ceramic capacitors |
| Dalian Dalicap | RF/microwave MLCCs and single-layer ceramic capacitors | High-frequency and low-loss requirements differ from low-voltage, ultra-high-capacitance decoupling |
Based on first-half 2026 reports; this is not a technology or market-share ranking. Nanjing Huiju New Materials is also an MLCC supplier, but recent dated product updates and AI high-capacitance delivery evidence were not established for this review.
Torch's commercial share of self-produced component revenue increased from 31% to 63% in the first half, a change that includes other component types. Hongda's multilayer ceramic-capacitor revenue was RMB 76.2871 million, down 24.40% year on year. These figures illustrate different business conditions within the wider ceramic-capacitor market.
High-reliability suppliers may bring quality systems and specialist processes; RF suppliers emphasise high-frequency loss and stability. Expansion into large-volume high-capacitance products still requires evidence on capacitance density, manufacturing economics and qualification for the intended customer application.
Global product development spans several locations in the power system
The international comparison includes Murata, Samsung Electro-Mechanics, TAIYO YUDEN, TDK and KYOCERA AVX. Taiwan-based suppliers include Yageo, Walsin Technology, Prosperity Dielectrics, Holy Stone and Darfon. KEMET is considered within Yageo rather than counted again as an independent group.
Murata's June 2026 securities report confirmed that a 1005M, 47 μF product had entered volume production. Samsung's December 2025 catalogue and April 2026 technical material listed a 0603, 100 μF, 2.5 V, X6S part. Compared with VIIYONG's 0805 entry above, Samsung's published specification uses a smaller imperial case code at the same nominal capacitance, voltage and temperature characteristic. This is a comparison of two specifications, not the manufacturers' entire portfolios or their effective capacitance in operation.
TAIYO YUDEN is also advancing substrate-embedded devices. Its December 2025 announcement described a metric 2012, 100 μF embedded MLCC, with production having started in November 2025, for IC power decoupling in information equipment including AI servers. Highly flat, precise external electrodes support reliable connection when embedded.
Figure 3. Manufacturer product photograph and specification table from TAIYO YUDEN's December 2025 announcement. The 2012 designation is a nominal metric case code; the table gives maximum dimensions. An embedded component is not a direct substitute for an ordinary surface-mount device.
TDK's September announcement addresses another power-system location: a 100 V, 10 μF, X7R low-resistance soft-termination MLCC for applications including 48 V AI-server systems, with production planned to begin that month. Voltage capability and resistance to mechanical stress are important here. Its function differs from 2.5 V-class high-capacitance decoupling near a processor.
Taiwan-based suppliers also target distinct requirements. Walsin's May investor material included high-capacitance, high-voltage resonant MLCC development. Holy Stone focused on low-loss, high-voltage NP0 products for server supplies, battery backup units and resonant circuits. Prosperity Dielectrics emphasised medium- and high-voltage, large-case, high-reliability products and dielectric materials. These are opportunities within AI infrastructure, with different capacitance and design requirements.
Yageo spans several passive-component categories, so its AI revenue share cannot be read as a high-capacitance MLCC share. Darfon also has other businesses, and its August material noted pressure from MLCC raw-material costs on gross margin. KYOCERA AVX remains part of the supplier landscape, without a claim here about specific new high-capacitance specifications or supply volumes.
Long-term supply arrangements add another competitive requirement. Samsung's AI-server MLCC contract disclosed in September covers the whole of 2027. Chinese manufacturers seeking broader adoption need sustained records of batch quality, change control and capacity assurance alongside product-specification progress.
Product mix can improve returns while expansion continues to absorb costs
EYANG's listed parent reported first-half MLCC revenue of approximately RMB 342 million, up 20.4%. The segment's gross margin rose from 16.5% to 20.3%. Management associated the improvement with a higher-end product mix and selling prices rising faster than average costs. These results cover the entire MLCC segment and cannot be attributed wholly to AI high-capacitance products.
Figure 4. Two separate first-half 2026 disclosures: the gross-margin change in EYANG's MLCC segment and the revenue change in Hongda's multilayer ceramic-capacitor business. Different products and customers prevent a direct comparison of company profitability.
Expansion-stage manufacturers face additional pressures. YUNZHONG's first-half revenue rose 53.68%, while its net loss attributable to shareholders was approximately RMB 91 million. Group revenue growth had not yet restored profitability, and its existing consumer-electronics business and MLCC investment require separate analysis. Lihexing reported that a June increase in some MLCC spot quotations briefly helped that month's profitability, but the benefit was short-lived and overall gross margin remained under pressure.
Fenghua and Three-Circle's reported third-quarter net-profit growth likewise belongs to consolidated businesses with other product lines. It is not a measure of AI high-capacitance MLCC margin.
As of early November 2026, China's supply options are broadening through new specifications, production progress and customer shipments. General-purpose order transfers and upgrades toward AI high-capacitance products are occurring together, but their beneficiaries and timing differ. The next commercial test is whether qualified parts can sustain larger deliveries and whether higher-end revenue can cover the additional manufacturing, research and depreciation costs.
