Most homelabbers can tell you their NAS’s total raw capacity down to the terabyte and argue at length about CPU generations, but ask what the whole rack costs to run for a month and you’ll get a shrug. That’s understandable, electricity is a background cost that never shows up as a single line item the way a drive purchase does, but a lab that runs 24/7/365 adds up in ways a lot of people never actually check. Here’s how to measure it for real, where the draw actually comes from, and what’s worth changing.

Stop guessing from the label

The number printed on a PSU or a spec sheet is the maximum rated output, not what the device actually pulls under normal use, and it’s frequently double or triple the real number for gear sitting mostly idle. A server with a 750W power supply might genuinely draw 60-90W at idle and only approach that 750W figure under a synthetic full-load stress test you’ll never actually run. Guessing from nameplate wattage is the single most common way people get their power-cost math wrong, almost always by overestimating.

The fix is cheap and simple: a plug-in power meter (the classic “Kill A Watt” style devices, or a smart plug with energy monitoring built in) sits between the wall and the device and reports actual watts drawn, in real time, over however long you leave it connected. For anything on a UPS, most mid-range and better UPS units also report live load wattage through their own web UI or management software, which is worth checking before buying a separate meter for gear that’s already behind one.

Measure over at least a full day, ideally a few days, not a single spot check. Idle draw, background scrub/backup jobs, and any scheduled tasks all shift the number, and a five-second reading tells you almost nothing about what the device actually costs over a billing cycle.

Where the watts actually go

In a typical home lab, the draw isn’t evenly spread. A few categories dominate:

Spinning drives. Each HDD typically draws somewhere in the 4-10W range at idle and noticeably more during seeks or a rebuild, and it’s rarely just one drive, a NAS with 8+ bays running mostly full spends a meaningful chunk of its total draw just keeping platters spinning. This is the category people underestimate most, because a single drive looks trivial and an array of them doesn’t feel like it should add up the way it does.

Older or enterprise-class hardware. Server-grade boards, redundant power supplies, and enterprise NICs and switches are built for datacenter power budgets, not home efficiency, and it shows in idle draw. A used enterprise switch with a wall of always-on transceivers or 10GBase-T ports (see the earlier piece on buying used enterprise gear for the buying-decision side of this) can idle at 50-100W or more just moving no traffic, where a modern consumer or prosumer switch handling the same port count might idle in the single digits to low tens of watts.

GPUs, even idle ones. A GPU sitting in a box for occasional transcoding or AI workloads still pulls meaningfully more at idle than no GPU at all, and some cards idle considerably higher than others depending on driver power states and whether anything is actively polling them.

Anything with active cooling running constantly. Fans are a small individual draw each, but a chassis full of them, especially small high-RPM server fans rather than large quiet case fans, adds up over a full month of continuous runtime, on top of the noise most of that gear also makes.

Doing the actual math

Once you have real watt readings, the conversion to cost is simple: watts times hours running, divided by 1000, gives you kilowatt-hours, and kilowatt-hours times your utility’s per-kWh rate (check your actual bill, not a national average, rates vary a lot by region and by time-of-use plan) gives you real cost. A device drawing a steady 80W and running 24/7 for a 30-day month works out to roughly 57.6 kWh for that month, and from there it’s just multiplying by whatever your utility actually charges per kWh.

Do this per device for anything running continuously, then total it up. The number that matters isn’t any single device, it’s the sum across everything that’s on 24/7, since that’s the actual monthly cost the lab adds to the household bill regardless of whether anything is being actively used at any given moment.

What’s actually worth changing

Consolidate before you optimize. The single biggest lever most people have is running fewer physical boxes doing more, rather than several older machines each idling separately. A handful of aging tower PCs repurposed as “extra homelab boxes” almost always costs more in combined idle draw than one modern, denser host running the same workloads as VMs or containers, even before counting the noise and rack space saved.

Right-size always-on switching gear. If you don’t actually need a full 10GbE-everywhere switch idling at high draw around the clock, a smaller or more efficient switch sized to what you actually use cuts a real, continuous cost. This is worth revisiting anytime port needs change, not just at purchase time.

Let drives spin down when it’s actually safe to. Spin-down on idle helps on drives that sit unused for long stretches, but think through the tradeoff first: constant spin-up/spin-down cycling adds mechanical wear, and drives that are part of an active array or a scrub schedule may never actually reach idle long enough for it to help. It’s a good fit for cold or archival storage, a worse fit for anything serving live reads regularly.

Check BIOS/UEFI power management. C-states, ASPM, and similar CPU power-scaling settings are frequently left at whatever a server board’s factory default was (sometimes tuned for consistent latency over efficiency), and enabling deeper idle states can meaningfully cut draw on hardware that spends most of its time waiting rather than computing. Worth a quick check on anything server-grade that’s been running on out-of-box settings.

Schedule what doesn’t need to run continuously. Dev/test VMs, occasional-use services, and anything that exists for convenience rather than 24/7 function are candidates for a shutdown/wake schedule instead of running around the clock. This doesn’t apply to anything load-bearing, but a lab tends to accumulate services that quietly run forever just because nobody turned them off.

What not to cut

Don’t apply any of this to backups, monitoring, or anything security-facing (a reverse proxy, a firewall, an alerting stack) just to trim watts. The whole point of measuring is to find the genuinely wasteful draw, mostly idle drives, oversized switching gear, and forgotten always-on services, not to talk yourself into a less reliable lab to save a few dollars a month. If a service exists specifically to catch problems or protect the rest of the stack, its uptime is worth more than its electricity cost.

Treat it as a real number, not a mystery

The habit that actually pays off isn’t a one-time audit, it’s checking in periodically, especially after adding new gear, since it’s easy to add “just one more” drive or box without noticing the cumulative draw creeping up. Measure what you actually have, do the real math against your actual utility rate, and you’ll usually find the answer is somewhere between “a rounding error” and “worth fixing,” never as scary as people assume before they check, and never as trivial as they assume after they stop checking.