Understanding Power Distribution in Data Centers (PDU, UPS, ATS)

A data center is fundamentally a power distribution and cooling machine. The servers inside are the payload, but the entire facility exists to deliver clean, reliable, uninterrupted power to those servers — and to remove the heat they generate. If power is interrupted even briefly, servers go down. In a hyperscale facility serving millions of users simultaneously, downtime is measured in millions of dollars per minute.

Understanding how power flows through a data center is essential knowledge for anyone working in electrical construction, commissioning, or operations. It is also the foundation for understanding why certain maintenance tasks require so much care — and why certain mistakes have catastrophic consequences.

Power starts at the utility grid. Large data centers negotiate directly with utility companies for dedicated power delivery — often 50–500 megawatts or more for a large campus. The utility delivers this power at high voltage (typically 115 kV to 345 kV, depending on the utility and the distance from the power source).

That high-voltage power arrives at an on-site or nearby substation. The substation contains step-down transformers that reduce the voltage to a medium-voltage level that can be used on the campus — typically 12.47 kV or 34.5 kV. The substation also contains switchgear that allows the utility feed to be switched, isolated, and routed.

Most tier 3 and tier 4 data centers are fed by at least two independent utility feeds from different substations. This redundancy means that if one utility feed fails — a downed transmission line, a substation fault — the other feed can carry the entire load. The switching between feeds happens at the switchgear level, often automatically within milliseconds.

From the main substation, medium-voltage power runs through underground conduit to the data center building. Inside the building, it enters the main switchgear room — a large room filled with metal-enclosed switchgear cabinets that route and isolate power to different systems.

The medium-voltage switchgear feeds main transformers, which step the voltage down again to the utilization voltage — typically 480 volts (three-phase) in the US. This 480V power is what most of the building's electrical systems run on.

The switchgear and transformer rooms are among the most restricted areas in a data center. Access requires documented training, proper PPE (arc flash suit, face shield, insulated gloves), and often a two-person rule — you cannot work alone in the high-voltage rooms. The arc flash energy available at medium-voltage equipment is lethal at distances of several feet.

Here is where data center power gets interesting. Even with redundant utility feeds, utility power is not perfect — it can have brief dips, sags, transients, or momentary outages. None of that is acceptable for servers.

UPS (Uninterruptible Power Supply) systems solve this problem. A UPS sits between the utility power (after the transformer) and the protected loads. It continuously charges a battery bank while simultaneously passing conditioned power through to the loads. If utility power dips or fails, the UPS switches seamlessly to battery — so fast (measured in milliseconds) that servers never notice the transition.

Large data centers use industrial UPS systems, not the small boxes you might plug a computer into at home. Rack-scale UPS units in hyperscale facilities might be 500 kVA to 2,500 kVA each. A large data hall might have a dozen such units arranged in parallel, with each unit able to carry the full load if others fail — this is N+1 or 2N redundancy.

The battery banks behind these UPS systems are enormous. Battery rooms in large data centers contain hundreds of battery strings — sealed lead-acid, lithium-ion, or newer chemistries — capable of delivering full power to the critical loads for 5–20 minutes. That is enough time for the backup generators to start and reach full load.

Five to twenty minutes of battery runtime is not enough to ride out an extended utility outage. Diesel generators are the answer. Every serious data center has backup generation capacity equal to 100% of its IT load — and most have more than that.

Generators on hyperscale campuses are large diesel units — typically 1 MW to 3 MW each, with large facilities having dozens of them. They are housed in dedicated generator enclosures outside or in separate generator buildings, with fuel tanks sized for 24–72 hours of runtime under full load.

When utility power fails, the UPS batteries hold the load while the generators start (typically 10–30 seconds to reach full speed and frequency stability). Once the generators are ready, the Automatic Transfer Switch (ATS) — explained next — switches the load to generator power. The batteries stop discharging and the facility runs on generator power until utility is restored or fuel runs out.

Generator testing is a significant part of data center operations. Monthly generator exercises, annual load bank tests, and integrated system tests (where utility power is actually cut to verify the full chain works) require specialized knowledge and careful coordination. Getting it wrong can cause the very outage you are testing against.

An Automatic Transfer Switch (ATS) is a device that monitors two power sources — typically normal utility power and emergency generator power — and automatically transfers the load from one to the other when conditions require it.

When utility power fails, the ATS detects the failure, sends a signal to start the generators, waits for the generators to reach stable voltage and frequency, and then transfers the load. When utility power is restored and has been stable for a programmed delay (typically 5–30 minutes, to avoid re-transferring during a momentary restoration), the ATS transfers back to utility and the generators cool down and shut off.

Data centers often have multiple ATS units — one per electrical panel or distribution board — to provide more granular control and redundancy. A static transfer switch (STS) is a solid-state version with transfer times measured in microseconds rather than milliseconds, used for the most sensitive loads.

After the UPS and ATS, power flows to Power Distribution Units — the PDUs. A PDU in a data center context is not the simple power strip you might think of. Floor-standing data center PDUs are rack-height or larger electrical distribution cabinets that receive 480V three-phase power and distribute it to multiple smaller circuits.

Each PDU might feed 20–40 individual 208V/30A or 208V/20A circuits, which in turn connect to the power strips (also called rack PDUs or rPDUs) mounted inside server racks.

The path is: utility → transformer → UPS → branch panel → floor PDU → rack PDU → server power supply. Multiple redundant paths exist at every level — a server might have power supplies connected to two different PDUs fed from two different UPS units. If either path fails, the server keeps running on the other.

Knowing the power chain is not just an intellectual exercise. For electricians doing construction or commissioning work, it determines safe work boundaries: which systems are energized, what the isolation points are, and what the consequences of an error are.

For operations technicians, it is the foundation for every maintenance task and troubleshooting scenario. When a circuit trips, understanding the power chain tells you which system to check first. When a UPS alarms, knowing what it protects tells you the urgency of the response.

The roles that pay $95K–$165K in data centers — critical facilities engineers, commissioning agents, controls technicians — all require this systems-level understanding. You do not need an engineering degree to learn it, but you do need to actively seek it out. Reading one-line electrical diagrams (the schematic maps of a facility's power distribution), watching what happens during generator tests, and asking experienced CFEs and commissioning agents to walk you through the facility logic are how you build this knowledge on the job.

Power distribution literacy is one of the clearest differentiators between a technician who tops out at $65K and one who builds toward $120K. It is learnable. The resources are the facility drawings, the systems, and the experienced people around you — if you pay attention.