From the Substation to the Server Hall

The National Electrical Code, published by the National Fire Protection Association as NFPA 70 and adopted into law by most American states, contains an article specifically for rooms that house information technology equipment. Its existence is a reminder that a data center, whatever else it is, is an electrical installation first, subject to the same physics and the same fire codes as a factory. This lesson follows the path electricity takes from the utility's wires to a server, one stage at a time.

Large utility customers are often served at high voltage. Voltage is the electrical pressure that pushes current through a wire. Higher voltage lets the same amount of power travel with less current, and less current means less energy lost as heat in the wires. That is why long-distance transmission lines run at tens or hundreds of thousands of volts. A large campus may therefore take its power directly from a transmission line through its own on-site substation, a fenced yard of switches, protective devices and large transformers.

A transformer changes alternating current, or AC, from one voltage to another. The substation steps transmission voltage down to what engineers call medium voltage, a range that in American practice typically runs from a few thousand volts to a few tens of thousands. A standard approved by the American National Standards Institute, known as ANSI C84.1, defines the preferred voltage ranges utilities and equipment makers design around. Medium voltage is used to carry power around the campus to each building, because it still keeps current, and therefore cable size, manageable.

Inside or beside each building, more transformers step medium voltage down to low voltage, in the United States commonly 480 volts for large equipment. Here the power passes through switchgear, which is the set of large switches and circuit breakers that route electricity and isolate faults. Switchgear decides which source feeds which load: the utility in normal operation, backup generators when the utility fails.

Next comes the uninterruptible power supply, or UPS. Its job is to bridge the gap between a utility failure and the moment generators take over, and to clean up disturbances in the incoming power. A traditional double-conversion UPS converts incoming AC to direct current, or DC, stores energy in batteries or a flywheel, and converts it back to AC for the load. Every conversion wastes a little energy as heat, which is one reason engineers have long looked for ways to skip steps.

From the UPS, power flows to power distribution units, or PDUs, which divide it into circuits, and then along busways or cables to the racks. A rack is the tall cabinet that holds the servers. Inside the rack, power supply units convert AC to the low-voltage DC that chips actually use. By the time electricity reaches a processor it may have been transformed or converted half a dozen times since it left the transmission line.

Not all of the electricity entering the building reaches the servers. Some is lost in conversion, and much of the rest runs the cooling systems that carry heat away. The industry's standard measure of this overhead is power usage effectiveness, or PUE, a ratio defined by the industry group The Green Grid: total facility energy divided by the energy used by the computing equipment. A PUE of 1.0 would mean no overhead at all. The Uptime Institute, which surveys operators every year, reported an average of about 1.54 in its 2025 survey and has noted that the figure has barely moved for several years. Google publishes a fleet figure well below that, which illustrates how much depends on design, scale and climate.

Two cautions apply to PUE. It measures overhead, not usefulness, so a building full of idle servers can post a fine number. And it is reported by operators themselves, often as an annual average that hides seasonal swings. Treat it as a question to ask, not a verdict.

Faraday's read is that the power chain is best understood as a series of handoffs, each with a cost in efficiency, space and money, and each a place where something can fail. Every design choice in a data center's electrical system is a decision about how many handoffs to keep and how to protect them. The next lesson looks at the protection.