When a data center PSU moves to a 48 V or 800 V bus, the aluminum electrolytic capacitors stop being interchangeable. The bulk capacitor, the hold-up capacitor and the output capacitor all sit in different parts of the power chain, and each one is selected on a different primary parameter. Get the wrong priority and the part passes the datasheet check but fails in the rack. This article is the selection logic behind the spec, not the spec sheet itself.

Why 48 V and 800 V Change What You Optimize For

We covered the architecture shift in our application overview, so here we only need the part that changes the capacitor decision. At 48 V the bus current is low enough that ripple current is manageable, but hold-up time over a short power interruption drives capacitance demand. At 800 V the opposite is true: voltage derating becomes the gate, and ripple current can be handled with multiple lower-voltage parts in series.

In short, the higher-voltage design trades a big single bulk capacitor for a string of series-connected cans with balancing, and it trades raw capacitance for voltage margin. That single shift is why a datasheet that looks perfect for a 12 V brick is wrong for an 800 V bus.

Three Capacitor Jobs, Three Different Selection Priorities

Inside a data center PSU the aluminum electrolytics are not one component; they are three, and each is chosen on a different axis.

• Bulk capacitor — sits on the rectified DC bus. Selected on ripple current rating and voltage rating, because it carries the bulk of the AC ripple from the front-end.
• Hold-up capacitor — stores energy to ride through a mains dropout. Selected on capacitance and ripple current, because the required energy is a direct function of C and the discharge window.
• Output capacitor — smooths the converter output. Selected on ESR and ripple current, because output ripple voltage is mainly ESR × ripple current.

Selection is not “pick the biggest capacitance that fits.” It is picking the number that the position actually needs. For the bulk and hold-up positions it is capacitance and ripple current; for the output it is ESR. The table below summarizes the priority.

Three Numbers That Do Most of the Decision

If you narrow every data center electrolytic to three numbers, you can turn a wall of datasheet rows into a short list.

• Ripple current at the operating frequency — the part is spec’d at 120 Hz (or 100 kHz for low-impedance types). Multiply by the frequency correction factor, never compare a 120 Hz number directly to a switching-frequency requirement. This is the leading cause of overheated parts.
• Endurance at the worst-case temperature — rated hours are quoted at 105 °C (or 85 °C). The classic rule of thumb doubles life for every 10 °C of temperature reduction, so a part rated 2,000 h at 105 °C is a different animal at 65 °C. Check the actual operating temperature, not the ambient.
• ESR and its change over the part’s life — ESR rises as the electrolyte dries. For the output position, plan the ripple with a worst-case ESR, not a fresh-from-stock value, or the end-of-life ripple will be higher than your simulation.

Ripple current and temperature are inseparable: the allowed ripple on the sheet assumes a maximum core temperature, and exceeding it shortens life faster than capacitor value drops.

Voltage Derating and Temperature: The Two Rules Engineers Forget

Two rules consistently trip up data center designs more than the arithmetic.

• Derate the voltage. Running an electrolytic at its full rated voltage continuously shortens life. A common design practice is to keep the working voltage comfortably below the rated voltage—on an 800 V bus that usually means using parts rated well above 450 V, often a series string, rather than squeezing a single 450 V part to its limit.
• Series strings need voltage balancing. Two capacitors in series do not automatically share the voltage equally; leakage current differences force one to take more. Add balancing resistors sized for the worst-case leakage, and de-rate for the fact that the effective capacitance of a series pair is halved.

For the range of techniques across the whole power chain, our snap-in electrolytic guides and screw-terminal guides show the form factors typically used for bulk and hold-up duties.

A Fast Walkthrough: Bulk Capacitance for a 48 V Bus

Rather than a generic formula, here is the order of operations for the position that challenges most engineers: the bulk capacitor.

• Start from the allowed ripple voltage on the bus. That sets the required minimum capacitance, roughly C ≥ I ÷ (f × ΔV), at the ripple frequency.
• Then size the ripple current. The part must carry the RMS ripple at that frequency, after applying the frequency correction factor.
• Then check temperature: confirm the core stays within the rated limit at the worst-case ambient plus local heating.
• Only then pick the form factor. For a 48 V bus, the capacitance and ripple demands often push toward snap-in or screw-terminal parts rather than small radial cells.

At 800 V the flow is the same but the voltage check comes first and the capacitance check is usually satisfied more easily. The series string and balancing are what you are really buying.

Common Selection Mistakes in Data Center PSUs

• Comparing a 120 Hz ripple rating to a switching-frequency requirement directly.
• Using the ambient temperature instead of the worst-case core temperature for endurance.
• Assuming two series capacitors share voltage equally without balancing.
• Treating ESR as a constant and ignoring its rise over the part’s life.
• Picking the biggest capacitance because “more is safer,” when the position needs ripple current or ESR instead.

When a program needs a part rated for the ripple and life, Xuansn supplies radial, snap-in and screw-terminal aluminum electrolytic capacitors with the ripple-current, endurance and ESR data laid out for exactly this kind of selection.

Common Questions About Selecting Data Center Capacitors

Snap-in, screw-terminal or radial for a data center PSU?

It depends on the position and the ripple current. Radial parts are compact and fine for low-ripple output positions; snap-in and screw-terminal parts carry far more ripple current and are the usual choice for bulk and hold-up duty on a 48 V or higher bus. Match the form factor to the ripple, not to the capacitance alone.

How much voltage derating is enough for an 800 V bus?

There is no universal answer, as it depends on the ripple, the expected core temperature and the operating life. The safe starting position is to keep the continuous working voltage comfortably below the rated voltage and to use a series string where the bus exceeds a single part’s rating—then sizes the balancing for that string.

Does higher capacitance always buy more hold-up time?

Only up to a point. Hold-up energy scales with capacitance, but the usable energy depends on how far the voltage is allowed to drop in the ride-through window. Once you know the minimum bus voltage the downstream converter accepts, extra capacitance beyond that point is mostly cost and size with diminishing return.

This completes our data center power series. Start with the 48 V and 800 V application overview, then review the full aluminum electrolytic selection guide for the fundamentals.