Capacitors

AI-Server Power Makers Expand Snap-In Capacitor Capacity: Why Jianghai Is Betting on Aluminum Electrolytics, Hybrid Capacitors and MLPCs

Higher server power is creating different capacitor jobs at the high-voltage bus, intermediate outputs and load point. Jianghai is expanding three complementary technologies, but long-duration reliability—not nameplate capacity—will determine design wins.

Two very different capacitor formats perform different jobs inside an AI-server power system: high-voltage snap-in aluminum electrolytics rated around 450 V, and compact surface-mount capacitors designed for low-voltage rails. Their size and specifications are far apart, yet both are necessary for stable power delivery.

Jianghai's August 2026 interim report disclosed a new capacity plan. The company is building and upgrading automated lines for data-center snap-in capacitors while accelerating technology and capacity expansion for polymer-hybrid aluminum electrolytics and multilayer polymer capacitors, or MLPCs.

This is not a bet on one universal "AI capacitor." High-power server architectures need different technologies for bulk energy storage, ripple filtering and fast load response at different voltage levels. The central question is whether China-based capacitor suppliers now have the reliability evidence and production discipline required for broad server-BOM adoption.

1. Why snap-in aluminum electrolytics come first

Jianghai reported first-half 2026 revenue of RMB 3.151 billion, up 16.96% year on year. Traditional aluminum electrolytics contributed RMB 2.498 billion and remain the company's core business. Compared with its smaller MLPC and hybrid-capacitor operations, the technically mature and broadly applicable snap-in product is the most direct route to meaningful volume.

The familiar snap-in aluminum electrolytic takes its name from the formed terminals that lock into a PCB. In a server AC-DC supply, rectification and power-factor correction create a DC bus at several hundred volts. Snap-in capacitors provide bulk bus energy storage, suppress low-frequency ripple and maintain output briefly after input power is removed.

As AI-server power rises, the component faces higher ripple current and more demanding hold-up-time requirements. Increasing capacitance alone is not enough. ESR, internal temperature rise, lifetime, airflow and mechanical volume must be considered together. High-voltage, high-capacitance, long-life snap-in devices are therefore among the BOM categories that server-power manufacturers most urgently need to qualify at greater volume.

Jianghai had previously announced volume orders for server capacitors. The interim report adds evidence that its data-center snap-in line is being upgraded. Public information still does not identify the end customer, order size or dedicated data-center part numbers. Commercial deployment is real, but it is too early to describe Jianghai as a leading-volume supplier to the entire sector.

2. Three capacitor technologies, three positions in the power chain

Server filtering and energy storage behave more like a layered reservoir system than a single component choice.

At the high-voltage bus, a snap-in aluminum electrolytic provides the large capacitance and voltage rating needed for bulk storage and low-frequency stabilization. After DC/DC conversion lowers the voltage and raises the current, capacitor priorities shift toward high-frequency filtering, ripple suppression and fast response to CPU or GPU load steps.

Polymer-hybrid aluminum electrolytics combine a liquid electrolyte with conductive polymer. The technology balances capacitance, low ESR, ripple capability and long-term reliability and is well suited to low-voltage, high-current, high-temperature power supplies in communications and data-center equipment. Jianghai has not yet published a complete data-center hybrid series or confirmed design-specific BOM details.

MLPCs use a multilayer polymer structure to reduce ESR and ESL for low-voltage, high-current, high-transient loads. Jianghai's HPA family spans 2 V to 25 V and 6.8 uF to 560 uF and can be used around DC/DC converters, switch-mode power supplies and processor-support rails.

These technologies are complementary rather than interchangeable. Snap-in devices provide bulk energy storage, polymer-hybrid devices balance performance and lifetime, and MLPCs sit closer to the load for rapid transient response.

Typical capacitor positions in an AI-server power chain

3. Meeting the headline specification is only the entrance test

Many China-made capacitors now approach international competitors on basic published parameters. The more difficult part of supplier substitution is stable behavior over years of complex data-center operation.

Snap-in qualification is demanding. Capacitance and rated voltage are only baseline values. Ripple-current capability must be derated for operating frequency, ambient temperature, airflow and target system life. Two parts with similar appearance and dimensions can produce very different full-load temperatures because of internal construction, ESR and thermal resistance, directly changing expected service life.

MLPC and hybrid devices cannot be selected on "low ESR" alone. Before use in a server supply, they need surge-current, ripple-voltage, reflow, temperature-humidity, mechanical-stress and lot-consistency testing. Their transient response must also be checked against real processor load steps. Small changes in copper impedance, layout or loop design can erase an apparent component-level advantage.

Jianghai's HPA family is a mature catalog range, but public material does not identify the voltage, capacitance, package and reliability grades dedicated to specific server programs. Until customer-qualified part numbers are confirmed, the family should not be described as a standardized drop-in replacement for server designs.

Procurement may focus on price, while engineering carries the implementation cost: prototype debugging, thermal testing, EMC correction, lifetime estimation and production-line changeover. A successful move to a China-based source is a system capability that includes validation, delivery and risk control—not a comparison of data-sheet rows.

Conclusion

Jianghai's expansion of data-center snap-in capacity and its parallel work on hybrid capacitors and MLPCs show that AI-server demand is moving from market narrative to physical manufacturing investment. The opportunity is not one capacitor part number; it is a structural change across the power BOM.

The status labels still matter: capacity expansion is not the same as qualified volume production, and a volume order is not proof of stable, industry-wide supply.

Snap-in aluminum electrolytics are Jianghai's most direct path into high-power AI-server supplies. MLPCs are the strategic route closer to CPU and GPU loads. Future design wins will depend less on installed capacity than on long-duration reliability evidence, production consistency, complete technical documentation and dependable delivery.

Capacity sets the supply ceiling. Validation determines whether a component is allowed into the design.

Disclaimer: This article is based on public information and is for industry discussion only. It is not investment, procurement or part-selection advice. Confirm specifications, orders, customer relationships and supply status with the manufacturer, customer and authorized channel.

Signals referenced in this article

The supply movement behind this piece, as recorded in the data. Figures are point-in-time snapshots carrying the date they were captured — they may have moved since publication.