Rack Power and Space Calculator: How Many kW and U Does a Colocation Rack Need?

Colocation planning usually starts with “how many U do we need?”, yet the real limit of a rack is more often electrical power than space. This rack power calculator works out IT load from the number of servers and network/storage devices and their measured average draw, adds a growth reserve and returns U space and rack count. As a colocation power calculator it also shows the smallest circuit that can carry the load on each feed and the heat the rack will produce.

All calculations run in your browser; nothing you enter is sent to a server.

Why kW, not U, is usually the real limit

A rack rarely runs out of physical space before it runs out of power and cooling budget. Dozens of 1U servers fit into a 42U rack, but their draw can exceed its circuit.

In colocation, a rack is a bundle of space, power, cooling and connectivity (explained in our guide what is colocation). The answer to “how many kW per rack” decides which circuit is installed, how much heat must be removed and how many racks the load is split across.

What the calculator asks for

The tool works from server and device counts, average draw per device, U height, extra U, growth reserve, rack size and feed type.

• Servers: count (default 10), average draw per server (350 W) and U per server (2U).
• Network and storage devices: count (3), watts per device (150 W) and U per device (1U).
• Other U: patch panels, cable managers and KVM (3U).
• Growth reserve: headroom for future additions (20%).
• Rack size and feed: 42U or 47U; redundant A+B feeds (default) or a single feed.

Enter the measured average draw per server, not the power supply (PSU) nameplate rating.

How the calculation works

Design load is IT load plus the growth reserve. For space, growth applies only to servers and devices, not to patch panels.

The formulas: IT load = servers × W + devices × W; design load = IT load × (1 + growth). Used U = servers × U + devices × U + other U; planned U = (servers × U + devices × U) × (1 + growth) + other U, rounded up. Racks = planned U ÷ rack size, rounded up, and load per rack = design load ÷ racks.

Heat load is the design load in watts multiplied by 3.412, giving BTU per hour; one watt is roughly 3.412 BTU/h (British thermal unit, Wikipedia).

Choosing the circuit: single-phase vs three-phase

The calculator picks the smallest circuit that can carry the per-rack load. Usable capacity is voltage × current (also × √3 for three-phase), multiplied by an 80% continuous-load limit and a 0.95 power factor.

Circuit	Approx. usable capacity
16A 230V single-phase	≈ 2.8 kW
32A 230V single-phase	≈ 5.6 kW
16A 400V three-phase	≈ 8.4 kW
32A 400V three-phase	≈ 16.8 kW

If the per-rack load exceeds 16.8 kW, the calculator says the load must be split across more circuits or racks. At the same current, three-phase carries roughly three times the power of single-phase.

The 80% rule is common electrical design practice for continuous loads. Servers run around the clock, so rack load counts as continuous.

Example 1: the default values

With the defaults, today’s IT load is 3.95 kW and the design load with 20% growth is 4.74 kW. That fits in one 42U rack with a 32A 230V circuit on each of the A and B feeds.

The math: 10 × 350 W + 3 × 150 W = 3,950 W. Used space is 20 + 3 + 3 = 26U; planned space is 23 × 1.2 + 3 = 30.6, rounded up to 31U. Since 4.74 kW exceeds the 2.8 kW of a 16A circuit, the tool selects 32A 230V (5.6 kW usable), and heat output is about 16,173 BTU/h.

Example 2: a dense rack of 1U servers

Twenty 1U servers at 450 W plus four network devices (150 W, 1U each) still leave plenty of space, but single-phase power is no longer enough. The result is a 32A 400V three-phase circuit per feed.

IT load is 20 × 450 + 4 × 150 = 9,600 W, or 9.6 kW; with 20% growth the design load is 11.52 kW. Used space is 20 + 4 + 3 = 27U and planned space is 24 × 1.2 + 3 = 31.8, rounded up to 32U, so one 42U rack is enough. Because 11.52 kW exceeds the 8.4 kW of a 16A three-phase circuit, each of the A and B feeds needs 32A 400V (≈16.8 kW), and heat output is about 39,306 BTU/h.

The two examples differ by just 1U, yet power differs by about 2.4 times. A rack space calculator that only counts U misses this.

Nameplate rating vs measured draw

A PSU nameplate shows the maximum the power supply can deliver; what the server actually draws is usually well below that.

Real draw can be read from smart PDU outlet or feed readings, iLO or iDRAC power history, or the UPS panel. Measure for at least a week, including month-end, backups and batch jobs. Enter the average into the calculator and use the peaks to check that the growth reserve is large enough.

Why each A+B feed must carry the full load

With A+B feeds, if one feed fails the whole rack keeps running on the other. So each feed is sized for the full rack load, not half of it.

Normally, dual-PSU servers share load across both feeds and the PDU shows each at about half. If circuits are sized to half the total, a failure or planned maintenance on one feed overloads the other. A single feed is simpler, but the rack stops completely when that feed is lost.

Heat and cooling

Nearly every watt going into a rack comes out as heat, so high-density racks need cooling planned in advance.

Standard hot and cold aisles usually suit low-density racks. At higher densities, rack placement, airflow and interaction with neighboring racks should be discussed at the quote stage, and the calculator’s BTU/h output gives both sides a shared reference.

Growth reserve and U planning

The growth reserve is power and space set aside for hardware added after the rack goes live. The default 20% is a sensible starting point for near-term additions.

Close unused U with blanking panels so hot exhaust air does not recirculate to server intakes. Cable managers matter for both maintenance and airflow, and belong in the “other U” field.

How GPU servers change the picture

GPU servers draw far more power per chassis than standard servers. A few can push a rack toward the three-phase limit.

The calculator then often splits the load across racks or circuits. For GPU memory sizing, see our GPU VRAM calculator; for power, rely on measurements under load rather than vendor maximums.

Common rack power planning mistakes

1. Sizing on nameplate: the PSU label shows a ceiling, not real draw, and leads to oversized circuits.
2. Sizing A+B circuits for half the load: when one feed fails, the other cannot carry the rack and redundancy is lost.
3. Counting only U: the space looks fine, but power and heat limits are reached much earlier.

Colocation racks at MAV Cloud

MAV Cloud’s colocation and rack service runs on Equinix data center infrastructure in Istanbul. Power allocation per rack, feed layout and circuit type are clarified at the quote stage based on your calculated needs; options are on the colocation rack rental page.

MAV Cloud’s service level agreement sets first-response targets of 15 minutes for critical, 1 hour for high, 4 hours for medium and 1 business day for low priority.

MAV Cloud is certified to ISO/IEC 27001, ISO/IEC 27701, ISO 22301, ISO/IEC 20000 and ISO 9001 (see our certifications).

Frequently Asked Questions

How many kW does a colocation rack need?

It depends on the measured draw of the hardware inside. The calculator’s default setup gives a 4.74 kW design load, while twenty 450 W 1U servers in the same rack reach 11.52 kW.

Why do A+B circuits look twice as large as needed?

Each feed is sized to carry the whole rack alone if the other feed fails. In normal operation the load is shared, so each feed uses only part of its capacity.

Should I choose single-phase or three-phase power?

Up to about 5.6 kW per rack, single-phase circuits are usually enough. Above that, three-phase is needed, and loads above 16.8 kW are split across more circuits or racks.

How accurate is the calculator?

It is a planning estimate based on the averages you enter; the actual power allocation is agreed at the quote stage.

If the setup is already clear, the quote link on the result screen is the fastest route.