# Rectifier
A rectifier converts AC (alternating current) to DC (direct current). In data center power architecture, this is the critical conversion step between grid power and what servers actually consume.
**The basic physics:**
AC power alternates direction 50 or 60 times per second (depending on grid). Electronics need steady DC. A rectifier uses semiconductor diodes (or more commonly, active switching transistors in modern designs) to convert the oscillating AC waveform into a stable DC output. Every electronic device does this somewhere — the question in data center design is *where* in the chain you do it, and at what voltage.
**Traditional approach — PSU per server:**
Each server has its own internal power supply unit (PSU) containing a rectifier. Grid AC reaches the server, and the PSU converts it to 12V DC for the motherboard and components. A typical 42U rack might have 20+ servers, each doing its own AC-DC conversion independently. That's 20+ rectifiers, each with its own efficiency losses, heat output, and failure probability.
**OCP rack-level approach — centralised power shelf:**
The Open Compute Project moved the rectifier out of individual servers and into a shared "power shelf" at the top or bottom of the rack. This shelf contains high-efficiency rectifier modules that convert facility AC to 48V DC, which is then distributed to all servers via a copper bus bar. Each server only needs a simple, cheap 48V → 12V DC-DC converter.
A typical OCP v3 power shelf is 1U and delivers 15-18kW. Multiple shelves can be paralleled. Modules are hot-swappable — you can add capacity without powering down the rack.
**Why 48V specifically:**
It's about current. Power = Voltage × Current. For a given power level, quadrupling the voltage (12V → 48V) cuts the current by 4×. Since resistive losses scale with the *square* of current (I²R), you get a 16× reduction in distribution losses. That's why Google, who pioneered this at OCP in 2016, reported ~30% energy efficiency improvement from the conversion.
48V is also below the ~60V safety threshold in most jurisdictions, so it doesn't require the same level of electrical safety certification as higher voltages. This matters for maintenance — technicians can work on 48V bus bars without high-voltage qualifications.
**Efficiency comparison:**
- Traditional per-server PSU: ~87-90% efficient (each conversion)
- 12V rack-level PSU: ~7 percentage points better than traditional (~33% loss reduction)
- 48V rack-level PSU: ~1 additional percentage point over 12V rack-level (~7% further loss reduction)
These seem like small numbers until you multiply by thousands of racks running 24/7. At data center scale, 1-2% efficiency difference is millions in electricity costs annually.
**What's coming next — 800V DC:**
As AI racks push toward 100kW+, even 48V hits practical limits. Carrying 2,500A at 48V requires massive copper bus bars. The emerging architecture (pushed by Nvidia among others) uses 800V DC distribution within the rack, with a 16:1 DC-DC converter stepping down to 48V at the server. This reuses the existing 48V server ecosystem while dramatically reducing distribution copper and losses.
**Where Delta applies this:** Their MDC containers use rack-level rectifiers feeding [[48V DC and OCP Bus Bar]] architecture as the standard production design — not individual server PSUs.
---
See also: [[48V DC and OCP Bus Bar]] | [[Power Usage Effectiveness - PUE]] | [[UPS Autonomy Philosophy — Generator Bridge]]