Home server or VPS: the electricity maths

A home server is quiet, small and "free to run" – until the annual electricity statement arrives and disagrees. This guide gives you the arithmetic to settle the question with your own numbers.

The short answer

Multiply average power in watts by 8,760 hours, divide by 1,000, and you have the year's kilowatt-hours. Multiply by your price per kilowatt-hour and you have euros. At the German household average of 37.0 ct/kWh that BDEW's analysis of August 2026 reports for 2026 so far, one watt around the clock costs about €3.24 per year [1]. Whether a home server beats a rented VPS depends on four things: the average draw (not the peak), your own tariff, whether you already own the hardware, and what else the machine has to do. We give no blanket verdict. The examples below are assumptions, labelled as such, and the break-even formula lets you replace them with your numbers.

How it works

1. The formula. Energy in kWh = power in W × hours ÷ 1,000, and cost = energy × price per kWh [5]. A year has 8,760 hours, so a constant 1 W is 8.76 kWh. Our energy-cost guide covers the basics and the limits of this arithmetic. Here we use it for a decision.
2. Choose the price on purpose. BDEW's analysis of August 2026 puts the average household price for 2026 so far at 37.0 ct/kWh. That figure covers new-customer tariffs and basic-supply tariffs for 3,500 kWh a year, with the standing charge included proportionally [1][2]. Destatis reports 37.97 ct/kWh for the first half of 2026, as an average over all contracts that includes working and standing charges [3]. Its table splits this by consumption class: 41.65 ct/kWh for households using 1,000 to under 2,500 kWh, and 36.18 ct/kWh for 2,500 to under 5,000 kWh [4]. Low users pay more per kilowatt-hour because the standing charge is spread over fewer units. For a decision, your own bill is better than any average: a server adds to your consumption, but it does not add a second standing charge. The marginal price is therefore closer to your working price (Arbeitspreis) in €/kWh. That is our reasoning, not a figure from the sources. We use 37.0 ct/kWh below because it is documented and dated.
3. Idle, load and the average. Nameplate wattages and peak values overstate what a lightly used server costs. What counts is the time-weighted average: the sum of power × duration for each phase, divided by the total time. Published measurements show how wide the spread is. ServeTheHome measured used business mini PCs from 2020 and 2021 at about 11 to 13 W idle and 51 to 54 W under maximum stress, on 120 V power [8][9]. The Raspberry Pi team states that a Raspberry Pi 5 peaks around 12 W under a "power virus" workload, against 8 W for a Raspberry Pi 4 [10]. These are old or different devices from yours, so they show the shape of the problem rather than your result. Suppose, as an assumption, that a machine idles at 12 W for 20 hours a day and works at 40 W for 4 hours. Its average is (12 × 20 + 40 × 4) ÷ 24 ≈ 16.7 W. That is 146 kWh a year, or €54.02 at 37.0 ct/kWh.
4. Measure instead of guessing. Put a plug-in meter between wall and device and read the kWh counter after several days, not the momentary watts. The Verbraucherzentrale describes this method [5]. Pay attention to the meter. Stiftung Warentest notes that some models only recognise power above 2 W [6], and a TH Köln study found that household energy loggers above €20 usually reached 1 to 2 percent accuracy, while power below 1.5 W could not be measured with them at all [7]. The full method, with boundaries and protocol, is in our measurement guides. We do not repeat it here.
5. Count the extras at the wall. Everything that stays on because of the server belongs in the sum: the router and switch it needs, external disks, and a UPS. A UPS loses energy in two ways. Proportional losses grow with the load. Fixed losses stay the same whatever you connect, so at low load they take a larger share of the energy, and efficiency drops below about half load and sharply below 30 percent [11]. ENERGY STAR's certification database lists one certified 1,800 W double-conversion model drawing 30.62 W from the wall at 0 percent load [12]. That is a data-centre-class example, not a typical homelab UPS. It shows why you should measure your own unit, at the wall, upstream of the UPS. Our UPS page explains sizing and shutdown behaviour.
6. The VPS side of the sum. Take the monthly price × 12, and read the small print: the first-term promotion versus the renewal price, VAT, the billing period and add-ons. Contabo's page showed, on 24 Oct 2026, the Cloud VPS 4 (4 vCPU cores, 8 GB RAM, 100 GB SSD) at €4.40 per month excluding VAT for the first 24 months [13]. The VAT-inclusive figure on that page depends on the country selected, so use the one your checkout shows. Hostinger's German page listed the KVM 2 plan (2 vCPU cores, 8 GB RAM, 100 GB NVMe) at €7.79 per month in a Black Friday promotion, with a renewal price of €14.99 per month "for 2 years"; the page does not state the VAT treatment [14]. Hetzner's cloud page lists plans from €5.99 per month excluding VAT and marks some plan lines as currently unavailable, so check the order page [15]. A VPS is billed while it exists, even when it is switched off [16]. That removes one trick from the comparison: you cannot "save" by powering a rented server down, but you can by switching off your own.
7. Break-even. The break-even average power is the annual VPS cost ÷ (8.76 × price per kWh in €). Electricity is a gross price including VAT, so compare it with a gross VPS price. For orientation we take 19 percent VAT. This is our calculation from BDEW's breakdown, which lists 5.91 ct of VAT on 31.09 ct net [1]. The plans differ greatly in resources, so only the price is compared. Assumed electricity price: 37.0 ct/kWh. The table below the list shows the results. Read it like this: a home server that averages more than the listed wattage costs more in electricity alone than that plan, before hardware, UPS, internet and your time. A server below it costs less in electricity. For a machine you already own, electricity is the only running cost. For one you still have to buy, add the price ÷ expected years of use. Money also leaves out things. The VPS comes with a data-centre connection and a public address, and Contabo states 99.9 percent uptime [13]. Your home gets power cuts and an upload line you share with the household. The home server brings local storage, GPU or tinkering freedom, and a LAN with low latency.

Break-even by VPS plan, at an assumed 37.0 ct/kWh:

A worked example

This is an example, not a report on our systems, and we publish no consumption figures for our own hardware here. Assume you started a plug-in meter on a Monday and read it a week later. The counter went from 12.40 to 15.20 kWh. All numbers are invented for the example.

awk 'BEGIN {
  kwh_week = 15.20 - 12.40;        # 2.80 kWh measured
  kwh_year = kwh_week * 365 / 7;   # about 146 kWh
  price    = 0.370;                # EUR per kWh, assumption (BDEW, Aug 2026)
  w_avg    = kwh_week * 1000 / (7 * 24);
  printf "average %.1f W, %.0f kWh, %.2f EUR per year\n", w_avg, kwh_year, kwh_year * price
}'

The output is "average 16.7 W, 146 kWh, 54.02 EUR per year". Compare that with the annual cost of the plan you are considering. At 16.7 W, this assumed machine sits slightly above the 16 W break-even of the Contabo plan excluding VAT, and below the 19 W with VAT assumed. It is also below the 29 W of the Hostinger promotion. The verdict flips with the VAT assumption, which is the honest result: the two are close, so the non-price factors decide.

Checklist

• Find your working price in €/kWh on your last bill, and note the date of the tariff.
• Measure the machine with a plug-in meter for at least a week and read the kWh counter.
• Calculate the average power from the counter, and do not use the peak or the nameplate value.
• Include the router, switch, disks and UPS that run because of the server.
• Look up the VPS renewal price, the VAT treatment and the billing period before you compare.
• Compute the break-even power with the formula and compare it with your measured average.
• Add hardware cost ÷ years of use if you still have to buy the machine.
• Decide about uptime, public reachability and the offsite backup separately from the price.
• Repeat the calculation when your tariff or the VPS price changes.