High-voltage snap-in capacitors: capacity and lifetime trade-offs in server power supplies
For server power designers assessing Chinese capacitor suppliers, recent announcements point to a more specialised supply base. Aishi previewed a compact high-voltage series in July 2026; Jianghai disclosed dedicated production-line work in August; YMIN published detailed product and temperature-test data in October. These developments sit alongside TDK's September expansion of compact snap-in families, including ratings up to 500 V in selected series.
The application is the high-voltage DC bus, where aluminium electrolytic capacitors provide filtering and short-term energy storage. This is a different job from that of low-voltage polymer capacitors and MLCCs on a motherboard. Increasing capacitance within a restricted enclosure is valuable only if ripple heating, service life and production consistency remain acceptable. Recent disclosures show progress across these requirements, although product announcements, qualification and delivered volume remain distinct milestones.
Product updates and production investment target compact high-voltage designs
Aishi's July KT announcement described small snap-in aluminium electrolytic capacitors intended for AI servers, HVDC and high-density UPS applications, including the high-voltage bus after the PFC boost stage. It identified a development direction rather than a record of volume shipments. Buyers still need subsequent specifications, customer-validation results and supply disclosures to assess commercial availability.
Jianghai's first-half 2026 report addressed the manufacturing side. The company disclosed new and upgraded fully automated snap-in lines for data centres and continued work on increasing etched-foil specific capacitance. This represents an adjustment to product and production requirements, rather than simply more output of general-purpose capacitors. The additional supply ultimately available will depend on commissioning and orders.
Figure 1. Supplier disclosures in 2026. A product preview, production-line work, a range expansion and sample-test publication represent different commercial stages; they are not equivalent measures of capacity or revenue.
TDK's September expansion covered compact families including B43645, B43657 and B43658. The latter two cover 420–500 V. The portfolio retains different endurance grades, reflecting the need to balance available space with reliability requirements rather than maximise nominal capacitance alone.
Together, these actions support growing demand for capacitance density in high-voltage server power supplies. They do not establish a shortage or a general price increase across all snap-in capacitors. Products serving appliances, industrial control and photovoltaics have their own supply-demand conditions.
Ripple heating and operating temperature constrain the benefit of extra capacitance
The energy requirement provides a useful starting point. Consider an illustrative 3 kW supply delivering output for 20 ms while its bus falls from 400 V to 300 V. At an assumed conversion efficiency of 95% during discharge, the theoretical total capacitance is approximately 1,800 μF. This excludes negative capacitance tolerance, ageing and other design margins. A 20 ms hold-up requirement is not universal across server supplies.
At unchanged hold-up time and bus-voltage limits, higher output power requires more stored energy. GaN or SiC switching devices can improve the conversion stage, but do not eliminate that storage requirement. Greater capacitance per unit volume can therefore free useful layout space.
Nominal capacitance is only part of the calculation. Ripple current produces losses in equivalent series resistance, or ESR, and raises internal temperature. Temperature in turn affects the ageing of liquid-electrolyte aluminium capacitors. A smaller package must still accommodate the required ripple current and heat flow.
YMIN's October 2026 IDC3 disclosure gives a concrete example: a 475 V, 1,670 μF device measuring 30 mm in diameter and 100 mm in axial length, with a rated ripple current of 4,260 mArms at 120 Hz and 105°C. Frequency and temperature belong with that current rating; comparison with a rating specified at higher frequency or lower temperature would require further work.
The same publication reported five samples tested over temperature. At 120 Hz, average ESR rose from 63.96 mΩ at 20°C to 1,135.86 mΩ at −40°C. The result shows why room-temperature data cannot describe cold operation. It is a small-sample measurement, not a guaranteed value for every production unit or a system-lifetime validation.
Figure 2. Original YMIN ESR test sheet, published in October 2026 and dated 9 September 2026. The sheet covers five IDC3 475 V / 1,670 μF samples, 30 × 100 mm, at 120 Hz. Reading guide: the mean ESR is 63.96 mΩ at 20°C and 1,135.86 mΩ at −40°C. Original labels and measurements are preserved; this is not a supplier ranking.
Mechanical interpretation matters too. The 100 mm dimension is axial length, not evidence that the part can stand upright inside a 1U power supply. Horizontal mounting still requires checks on diameter, retention, safety clearances and airflow. A smaller diameter alone does not establish a system-level space saving.
Endurance figures require similar care. Lelon’s 2026 catalogue specifies 2,000 hours at 105°C for the LSG series, explicitly including rated voltage and rated ripple current. Nippon Chemi-Con's April information lists 3,000 and 5,000 hours at 105°C for KHU and LHU respectively. These are tests for different families, not direct estimates of server operating years or a complete basis for selecting one over another.
Increasing capacitance can add energy margin, but losses and obstructed airflow may increase cooling and validation requirements. Component selection and measured temperature rise therefore need to be assessed together.
Chinese suppliers are competing through materials and repeatable manufacturing
The recent Chinese disclosures span finished products, upstream materials and production equipment. Their significance depends on which part of the supply chain each company is advancing.
Jianghai links its high-capacitance development to electrode foil, automation and manufacturing management. Its half-year report describes work on foil with high specific capacitance, strength and consistency, alongside automated data-centre lines. For wound capacitors, material properties and processing consistency affect saleable output. A line's ability to sustain qualified production matters more commercially than its planned capacity alone.
Aishi also operates across electrode foil and finished capacitors. Its half-year report identifies high-voltage snap-in, high-temperature/high-ripple and miniaturised products as development and promotion priorities. Industrial-control and new-energy products, including automotive and data-centre applications, accounted for 54% of principal-business revenue. This is a broad application grouping, not an AI-server revenue share and not revenue attributable to KT.
YMIN's recent material is more specific to product evaluation. Its October IDC3 article discusses oxide film, electrolyte, sealing and connection processes as well as temperature measurements. Such disclosure provides a basis for evaluation; completed integration into a customer's power platform remains a separate step. Temperature rise, hold-up performance and long-duration operation will influence adoption.
Dongyangguang adds the upstream perspective. Its first-half report describes development of laminated formed foil and capacitors using that foil for high-end applications including AI servers. Higher anode-material specific capacitance can enable smaller finished components, but winding, impregnation, sealing and ageing remain necessary manufacturing steps. This foil technology should not be confused with motherboard MLPC multilayer polymer capacitors.
Figure 3. Original page from TDK's December 2025 compact snap-in technical material, showing anode-foil and electrolyte development and a winding example. Related families were expanded in September 2026. The illustration does not measure a performance gap between suppliers.
International comparisons need the same level of detail. TDK highlights its own anode foil, oxide film and electrolyte development; Nippon Chemi-Con also identifies new electrode foil as a means of increasing high-voltage capacitance. Competition therefore extends to coordination between materials and finished-device manufacturing.
The available disclosures establish activity in these areas, but do not support a comprehensive technical ranking. A meaningful comparison would hold voltage, capacitance, dimensions, frequency and temperature constant while examining ripple capability, endurance, batch consistency and supply records.
Revenue growth and cash generation give different views of expansion
Jianghai reported approximately RMB 4.920 billion in revenue for the first nine months of 2026, up 19.50%, and RMB 624 million in net profit attributable to shareholders, up 16.71%. Aishi reported approximately RMB 3.532 billion and RMB 311 million respectively, with growth of 19.34% and 40.13%. Aishi's year-on-year comparisons use a retrospectively adjusted base.
These consolidated figures cover multiple products and end markets. They do not disclose the revenue or profitability of server high-voltage snap-in capacitors specifically.
Figure 4. Consolidated results for January–September 2026, in RMB million. Each metric has its own zero-based scale. Source: the companies' third-quarter reports. These are not snap-in capacitor segment results.
Cash generation moved differently. Jianghai's net operating cash inflow was approximately RMB 474 million, down 25.94%. Aishi's was approximately RMB 117 million, turning positive from the prior-year period. Profit growth alone cannot show how quickly new business converts into cash; collections and procurement outlays also matter.
Higher-capacitance server products may carry greater value per unit, while requiring more expensive materials, tighter screening, equipment upgrades and customer validation. The return depends on those costs and qualified output. Similarly, a delivery constraint on selected high-capacitance parts does not establish a shortage in general-purpose appliance or industrial-control products. Unverified price-increase claims cannot support an industry-wide profit estimate.
Jianghai's half-year report also notes uncertainty in the scale and timing of data-centre infrastructure investment, requiring coordination of buildings, equipment and materials capacity with demand. As of early November 2026, the clearest development is the allocation of more research and manufacturing resources to compact high-voltage products. Capacitance density helps a component enter the design shortlist; ripple performance, temperature and endurance determine its operating suitability; repeatable output and cash collection determine the economic return.
Prepared from public information for industry discussion. Product parameters, qualification and supply status remain subject to current manufacturer and customer information.
