How much does an always-on homelab cost per month?

An always-on homelab drawing 60 W at the outlet would use 43.2 kWh in a 30-day month. At an assumed $0.15/kWh, that is $6.48; at $0.30/kWh, it is $12.96. The useful answer starts with your average wall draw and electricity rate, not the power supply’s label or a GPU power limit.

The cost examples use assumed wall draw and tariffs; the GPU figures are measured on our bench and labeled as board power. Use the running-cost method below with your outlet readings.

For power protection, use the UPS capacity and graceful-shutdown guide. Size battery runtime and test shutdown separately from your monthly energy budget.

Shopping for a smaller host? Our mini-PC buying guide compares exact Beelink, GMKtec and Minisforum configurations and upgrade limits. Use outlet readings to compare their running costs; PSU and processor ratings describe different things.

We earn a commission if you buy through our links — it doesn’t change our picks. How we make money → As an Amazon Associate I earn from qualifying purchases.

What you’ll need: Use a compatible plug-in meter for the outlet measurement: HBN U167J for retained cumulative readings, or Kill A Watt P4400 for a standalone local readout. Choose the meter for your outlet and load.

Calculate monthly cost from watts and hours

Use two steps:

Energy (kWh) = average wall power (W) × operating hours ÷ 1000
Electricity cost = energy (kWh) × price per kWh

This follows the Australian Government’s running-cost method: convert watts to kilowatts, multiply by hours, then apply the electricity rate from your bill. If you already have accumulated kWh from a meter, use that energy reading directly.

For the table below, assume 24 hours a day for 30 days, or 720 hours. All prices are USD; both rates are illustrative. Each wattage is an assumed constant reading at the wall, not a claim that a particular machine achieves it.

Constant wall draw	kWh/month	At $0.15/kWh	At $0.30/kWh
20 W	14.4	$2.16	$4.32
60 W	43.2	$6.48	$12.96
120 W	86.4	$12.96	$25.92
200 W	144.0	$21.60	$43.20

Use the actual number of days when estimating a specific month.

Measure the whole system at the outlet

First define the measurement boundary. A meter feeding only the server measures that server’s electricity input. It does not automatically include separate storage, network equipment or cooling. A group of devices connected downstream of the meter creates a different boundary; record what is included before comparing results. Once you have measured the protected load, use our small-rack UPS comparison to choose the backup-power hardware.

For your measurement, record the date, hardware under test, meter model, elapsed measurement time, accumulated kWh and workload. Include the firmware and software settings that could affect the comparison. Check that the meter is appropriate for the outlet and load, and follow its instructions.

A snapshot of idle watts can help diagnose a change, but it does not describe a variable workload’s monthly energy use. The US Department of Energy advises measuring energy across a period and dividing by its duration when power fluctuates. Choose a period that represents the intended workload, including both idle time and active work. Keep that observation period with the result.

Wall readings already include the power supply’s conversion losses within that boundary. Do not add a second estimated PSU-loss allowance to measured outlet energy. Conversely, the PSU’s rated output is not a measurement of what the server continuously draws.

Combine idle and active hours without double-counting

An always-on machine need not draw the same power all day. Consider this hypothetical workload: 60 W for 20 idle hours and 180 W for 4 active hours per day. Replace those assumed hours with your workload mix.

Monthly energy = ((60 × 20) + (180 × 4)) × 30 ÷ 1000
               = 57.6 kWh

That costs $8.64 at $0.15/kWh, or $17.28 at $0.30/kWh. During active hours, 180 W replaces 60 W; it is not an extra 180 W on top of idle draw.

An equivalent calculation starts with the constant 60 W baseline of 43.2 kWh, then adds only the difference between active and idle draw during active hours: 14.4 kWh, costing $2.16 or $4.32 at the two assumed rates. This distinction matters when a workload’s quoted watts describe total system draw rather than an increase above idle.

Why an NVIDIA power cap is not your electricity bill

A power cap sets a ceiling; actual draw changes with the work. And NVIDIA describes NVML’s power reading as board power, which does not supply the complete server-at-the-outlet figure needed for the budget above.

Our recorded September bench notes illustrate the difference between a cap and draw. In the September 25, 2026 RTX 3060 Ti sweep, qwen2.5:7b Q4_K_M under Ollama 0.34.2 produced 75.5 tok/s at a 150 W cap, with 145.4 W mean GPU draw. At a 200 W cap, it produced 77.8 tok/s, with 193.4 W mean GPU draw.

The method used three 512-token generation runs per cap after a warm-up, with seed 1 and temperature 0. Driver telemetry was sampled every 250 ms; the reported load-draw mean uses samples above 40 W from runs two and three. Keep those board-power readings separate from the whole-system outlet watts used in the monthly table.

For cap selection and limitations, read the NVIDIA power-limit guide. To evaluate electricity savings, compare outlet energy for equivalent useful work. A slower configuration may run longer; comparing watts alone misses that trade-off. Price the outlet-energy difference after measuring it.

Use your tariff and state what the budget excludes

Replace both example rates with the applicable usage rate from your electricity bill. For time-dependent pricing, calculate energy cost for each priced interval and add the results; one flat rate may misrepresent the cost.

This guide estimates electricity consumption cost. Hardware purchases, internet access, separate equipment and cooling outside the meter boundary are excluded. Fixed charges and taxes are also excluded from the examples. When comparing an incremental server cost with a total bill, state which charges would actually change.

For the useful answer, pair your tariff with representative measured energy.

Check the result before buying replacement hardware

Start with the largest measured energy contribution, then change one setting or component at a time and repeat a comparable workload. Preserve the original readings and settings so that an apparent saving can be checked. Include useful output or completed work when comparing active workloads.

For scale, a hypothetical 10 W reduction maintained continuously saves 7.2 kWh over 30 days: $1.08 at $0.15/kWh or $2.16 at $0.30/kWh. Compare an upgrade’s purchase cost with savings you have measured.

For the two-GPU B450 build, the recorded performance figures describe the tested cards and workloads. Measure its whole-system outlet energy to turn those performance results into your own running-cost comparison.

If you need to buy a meter first, the plug-in power-meter buying guide compares the P4400 with HBN and Kasa alternatives. Choose between local cumulative readings and app history, and check whether the totals survive unplugging before planning the cost measurement.