The same copper carries roughly 1.73× more power on three-phase than single-phase at the same voltage and current per conductor — which is exactly why high-density GPU racks abandon single-phase almost entirely.
Single-phase power delivers energy through one alternating voltage waveform between two conductors (line and neutral, or two lines in a split-phase system). Three-phase power delivers energy through three separate voltage waveforms, each offset 120 electrical degrees from the others, sharing a common set of conductors. For the same conductor size and voltage, three-phase distribution can deliver roughly 73% more real power than single-phase — a direct consequence of how the phases combine — while also producing a more constant, ripple-free power delivery instead of the pulsing (sinusoidal, zero-crossing) power delivery inherent to single-phase. Every legacy enterprise rack could get by on single-phase; almost no modern high-density GPU rack can, and understanding why is a matter of watts per conductor, not just "bigger cable."
A single-phase waveform crosses zero volts twice every cycle, meaning instantaneous power delivery pulses rather than staying constant — for equipment like motors this causes torque ripple, and for any load it means the conductors must be sized for peak, not average, current. Three-phase power staggers three waveforms 120° apart specifically so that when one phase is near zero, the other two are well away from zero — the sum of instantaneous power across all three phases is constant, not pulsing. That constant-power property, combined with the √3 factor from how line-to-line voltage relates to line-to-neutral voltage in a three-phase system, is what lets three-phase distribution deliver roughly 1.73× more real power through the same conductor ampacity as single-phase at the same voltage. For a GPU rack drawing 30–80+ kW, that difference is the difference between a rack that needs one modestly sized three-phase circuit and one that would need multiple oversized single-phase circuits with more conductors, more terminations, and more PDU complexity.
No — a typical GPU server power supply is still a single-phase load internally; what changes with a three-phase rack PDU is that the rack's many single-phase server power supplies get distributed and balanced across the three phases (L1, L2, L3) rather than all pulling from one. A well-designed three-phase rack PDU deliberately spreads outlet groups roughly evenly across all three phases so that no single phase conductor is overloaded while the others sit idle — an unbalanced rack (say, most servers plugged into outlets fed by L1) can trip that phase's breaker even while L2 and L3 have spare capacity, a common troubleshooting trap for engineers unfamiliar with phase balancing. Getting this balance right, and confirming it during commissioning with a clamp meter on each phase, is a standard and necessary step when populating any three-phase-fed GPU rack.
Explains why three-phase power delivers roughly 1.73x more real power than single-phase through the same conductor ampacity, and why modern high-density AI/GPU racks are almost always fed by three-phase distribution even though each server's power supply is a single-phase load.
Single-phase power delivers energy through one alternating voltage waveform, typically on a line and neutral conductor pair (or two lines in a split-phase 120/240V residential-style system). Because the voltage waveform crosses zero twice per cycle, instantaneous power delivery pulses rather than remaining constant, and conductors must be sized to handle the peak, not average, current draw. Real power delivered is P = V × I × cos φ for the single conductor pair.
Three-phase power uses three separate voltage waveforms, each offset 120 electrical degrees from the others, typically delivered on three line conductors (plus sometimes a neutral, for a 4-wire wye system). Because the three waveforms are offset, at any given instant at least one phase is well away from its zero crossing, making the combined instantaneous power delivery constant rather than pulsing. Real power delivered is P = √3 × V(line-to-line) × I × cos φ, which works out to roughly 1.73 times more power than single-phase for the same conductor current and phase-to-neutral voltage.
A legacy enterprise rack drawing a few kilowatts is easily served by one or two single-phase 20A/120V or 30A/208V circuits. A modern AI/GPU rack drawing 30, 50, or over 100 kW would require an impractical number of separate single-phase circuits and conductors to deliver that much power; a single three-phase feeder at the same voltage and conductor ampacity delivers roughly 73% more power, dramatically reducing the number of circuits, conductors, and terminations needed per rack. This is why three-phase (commonly 208Y/120V or increasingly 415Y/240V in newer high-density designs) has become the standard rack-feed voltage for AI/GPU data halls.
Individual server power supplies remain single-phase loads even when fed from a three-phase rack PDU — the PDU's job is to distribute its many outlets across all three phases (L1, L2, L3) as evenly as possible. An unbalanced rack, where most load concentrates on one phase, can trip that phase's breaker or overheat that conductor even while the other two phases carry light load, a common and avoidable design and commissioning error.
A 415Y/240V three-phase system delivers roughly double the power of a 208Y/120V system at the same current, further reducing conductor size and circuit count for extreme-density racks. Some of the highest-density AI rack designs (100+ kW per rack) are moving toward higher-voltage three-phase or even DC distribution specifically to keep conductor sizes and connector counts manageable.
No — the power a server actually consumes is set by its workload and hardware, regardless of whether it's fed from a single-phase or three-phase circuit. Three-phase distribution is about efficiently delivering that power through the facility's wiring and PDUs, not about reducing the server's own draw.
Yes — this is exactly how most three-phase rack PDUs work: they accept a three-phase (plus neutral) feed and break it into groups of standard single-phase outlets (C13/C19), with each outlet group wired to one phase, spread across the PDU so the rack's total load balances reasonably evenly across all three phases.
A significant phase imbalance can be a genuine safety and reliability issue, not just an efficiency concern — an overloaded phase conductor can overheat and trip its breaker (or worse, in a fault condition, overheat insulation), and can also create neutral conductor overloading in certain configurations. Facility electrical engineers monitor and correct phase balance as a standard part of both design and ongoing operations.