You set up a NAS, a network-attached storage device that sits on your home network and runs apps without a separate PC, and loaded it up with Nextcloud, Immich, Jellyfin, Vaultwarden, Paperless-ngx and a handful of other containers. Now you are wondering what that is doing to your electricity bill. The honest answer is: it depends entirely on which apps are running and whether any of them are doing real work.
In short: The additional draw from 8-12 containers depends on the hardware and workload; calculate the annual cost from measured additional wattage and your actual tariff. Jellyfin transcoding and Immich ML scans are the cost outliers. Everything else is background noise.
What Docker Containers Actually Add to Your Power Draw
Docker, software that runs applications in isolated containers without affecting the core operating system, does not itself consume significant power. A container sitting idle consumes almost nothing. The power cost comes from what happens inside the container: CPU cycles for OCR, video transcoding, machine learning inference, and background sync processes.
Ten inactive containers may add little power on some systems, but idle overhead depends on the host hardware and on background polling, logging and storage activity; measure the specific NAS to establish it. The apps that break this pattern are Jellyfin during active transcoding, Immich during its nightly machine learning face-detection pass, and Paperless-ngx when ingesting a large document batch.
Base NAS Power Consumption by Hardware Tier
Your Docker stack sits on top of a baseline power draw that varies significantly by NAS hardware. ARM-based entry models draw a fraction of what Intel Core-based mid-range units do. Before counting container overhead, understand where your baseline sits.
NAS Power Draw by Hardware Tier (idle with drives spinning)
| ARM entry (DS223J) | Intel Celeron mid (DS425+) | Intel Core i3 (DS925+) | Intel Core i5 8-bay (DS1825+) | |
|---|---|---|---|---|
| Idle power (no containers) | Varies with drive count, drive model, configuration and system state; use measured wall power for the actual NAS. | Varies with drive count, drive model, configuration and system state; use measured wall power for the actual NAS. | Varies with drive count, drive model, configuration and system state; use measured wall power for the actual NAS. | Varies with drive count, drive model, configuration and system state; use measured wall power for the actual NAS. |
| 8-12 containers, all idle | 10-16W | 18-28W | 25-38W | 32-52W |
| Active sync (Nextcloud/Immich) | 14-22W | 25-38W | 35-55W | 45-70W |
| 1080p transcode (Jellyfin SW) | 25-40W | 40-65W | 55-80W | 70-100W |
| Annual cost idle (NSW ~$0.30/kWh) | $21-32 | $39-58 | $58-84 | $74-118 |
| Annual cost active mix (NSW) | $37-58 | $66-99 | $92-144 | $118-183 |
Hardware selection is the biggest lever in your power bill. The DS223j and DS925+ are not like-for-like hosts for this stack: the DS223j has 1GB RAM, while Immich specifies at least 6GB., which is $39-52 per year in NSW and $58-84 per year in South Australia. That difference compounds over a 5-year NAS lifespan into a meaningful cost gap.
Container Power Profiles: What Each App Actually Draws
Not all containers are equal. The apps people most commonly run on a home NAS sort into three power categories: background-idle, occasional-burst, and sustained-load. Knowing which category each app falls into tells you where your bill is actually coming from.
Docker Container Power Overhead on a Mid-Range 4-Bay NAS
| App | Idle overhead | Active overhead | Trigger | |
|---|---|---|---|---|
| Vaultwarden | Vaultwarden | +1-2W | +1-2W | Effectively constant. No heavy processing |
| Pi-hole | Pi-hole | +1-2W | +2-4W | DNS queries. Negligible at home scale |
| Nginx Proxy Manager | Nginx Proxy Manager | +1-2W | +2-5W | Active web traffic through the proxy |
| Uptime Kuma | Uptime Kuma | +1-2W | +1-2W | Polling checks. Minimal |
| Home Assistant | Home Assistant | +2-4W | +5-10W | Automations, integrations running |
| Paperless-ngx | Paperless-ngx | +1-3W | +8-20W | OCR during document ingestion |
| Nextcloud | Nextcloud | +1-3W | +5-15W | Active client sync, background tasks |
| Immich (idle) | Immich (idle) | +2-5W | +10-30W | Nightly ML face/object detection pass |
| Jellyfin (no stream) | Jellyfin (no stream) | +2-4W | +30-80W | Active software transcoding of 1080p |
| Whoami / lightweight utility | Lightweight utility | +0-1W | +0-1W | Near zero |
Jellyfin transcoding is the outlier. Software transcoding can materially increase CPU load and power use, but the increase depends on the NAS, source and output codecs, resolution, tone mapping and subtitle handling. Concurrent transcodes can raise power further, but the effect is not linear and should be measured on the specific NAS and media workload. If Jellyfin is in your stack and family members watch simultaneously, your power cost estimate needs a separate line item for streaming hours.
Real Annual Cost by Australian State
Australian residential electricity rates vary sharply by state. Flat residential usage rates vary by jurisdiction, distribution zone, retailer and plan; current regulator or standing-offer data should be compared on a like-for-like basis. These differences make the same NAS hardware cost nearly twice as much to run in Adelaide as in Canberra.
Annual Power Cost: 4-Bay NAS + 10 Containers (25W average draw)
| State | Approx rate (flat residential 2026) | Annual cost at 25W | |
|---|---|---|---|
| South Australia | South Australia | ~$0.44-0.48/kWh | $96-105/yr |
| Queensland | Queensland | ~$0.29-0.33/kWh | $64-72/yr |
| New South Wales | New South Wales | ~$0.28-0.32/kWh | $61-70/yr |
| Victoria | Victoria | ~$0.27-0.31/kWh | $59-68/yr |
| Western Australia | Western Australia | ~$0.29-0.33/kWh | $64-72/yr |
| Tasmania | Tasmania | ~$0.29-0.33/kWh | $64-72/yr |
| ACT | ACT | ~$0.26-0.30/kWh | $57-66/yr |
These figures assume a steady 25W average. Real usage will vary because Docker workloads are bursty, not constant. The NTKIT NAS Power Calculator lets you enter your actual state rate and estimated wattage to get a personalised annual figure. Use the table above to find your starting wattage estimate, then adjust for your specific app stack.
The Spiky Load Problem: Immich ML and Paperless OCR
Bursty containers are harder to cost than constant-draw ones. Immich runs machine-learning jobs such as smart search and face detection when assets are processed, including during initial library indexing and after new uploads. Initial library processing can create a sustained load, but its duration and power increase depend on the library, enabled jobs, processor and whether machine learning is local, hardware-accelerated or remote. Once the library is indexed, ongoing passes on new uploads are brief and much cheaper.
Paperless-ngx OCR is similar: ingesting a large batch of scanned documents uses real CPU for minutes to hours, then returns to near-idle. Neither of these apps represents a meaningful ongoing cost once the initial indexing pass is complete. Initial indexing can be a one-time spike, but Immich jobs recur on new assets and Paperless OCR recurs whenever new documents are ingested.
The practical takeaway: calculate your baseline power cost on idle wattage, add a line item for Jellyfin if household members actively stream, and treat everything else as noise. Without active video transcoding, container overhead still varies with the applications, background activity and NAS hardware; estimate it from measured wall power and the household's actual tariff.
How to Reduce Power Costs Without Cutting Your Stack
The most effective power-saving moves do not require removing any apps. Potential cost-saving measures include moving suitable tasks to cheaper tariff periods, enabling disk hibernation where workloads permit it, and selecting hardware that is efficient and capable of running the intended stack.
Disk hibernation is often left at its default setting and forgotten. Most NAS operating systems allow drives to spin down after 10-20 minutes of inactivity. In a pure-storage use case this works well. In a Docker-heavy stack, containers that poll the filesystem regularly (logging, monitoring, Nextcloud file scanning) can prevent drives from ever spinning down. Identifying wake events and reducing unnecessary polling may lower average draw by allowing drives to remain in hibernation longer; the saving depends on the drives and system configuration.
Hardware transcoding is the biggest single intervention for Jellyfin users. Compatible, correctly configured hardware acceleration can substantially reduce CPU load during transcoding, but actual power use depends on the GPU, codec pipeline, drivers and NAS implementation. Enabling hardware transcoding in Jellyfin settings, if your NAS CPU supports it, converts your highest-cost workload from sustained-load to near-idle.
Time-of-use tariffs: If your state offers a time-of-use electricity plan, If the household has a time-of-use plan, moving controllable workloads into that plan's cheaper periods may reduce their cost; verify the actual tariff windows and whether each application supports the required scheduling workflow. Synology DSM and QNAP QTS both support scheduled task execution at the OS level.
Australian Context: Rates, Tariffs and What to Watch
Australian residential electricity is billed under either a flat tariff (same rate per kWh at all hours) or a time-of-use tariff (peak/off-peak/shoulder rates). Households may be on flat, time-of-use or other tariff structures depending on their jurisdiction, meter, network and retailer plan. The AER's 2026-27 Default Market Offer covers New South Wales, South East Queensland and South Australia; figures for other jurisdictions must come from their relevant regulator or current retailer tariffs.
South Australian residents consistently pay the highest rates in the country. If you are in Adelaide and running a power-hungry stack, the hardware efficiency argument for an ARM-based NAS is stronger than anywhere else in Australia. Any efficiency payback should be calculated from measured wattage, the actual tariff and the purchase-price difference, while comparing models capable of running the same workload.
Check your latest bill for your actual supply charge and usage rate. The rates in this guide are representative benchmarks, not every retailer's current offer. The calculator provides approximate regional defaults; enter the usage rate from your latest bill for an accurate retailer- and plan-specific estimate.
Related reading: our NAS buyer's guide, our NAS power consumption guide, and our NAS explainer.
How much power does a NAS use running Docker containers 24/7?
Average power for a 4-bay NAS running containers depends on its processor, drives and workload and should be measured on the intended configuration; the 2-bay DS223j is not a like-for-like host for the stated Immich stack. Add 30-80W during active Jellyfin transcoding.
What is the most power-hungry Docker container to run on a NAS?
Jellyfin with software transcoding enabled is by far the highest power draw, adding 30-80W above idle during active streams. Immich ML scanning is the second highest but runs only in bursts during initial indexing and nightly passes on new photos. Everything else, including Vaultwarden, Pi-hole, Nextcloud, and Home Assistant, adds 1-5W at idle.
How much does it cost to run a self-hosted NAS stack in Australia per year?
For a mid-range 4-bay NAS drawing an average of 25W with 10 containers, annual power cost ranges from $57 in the ACT to $105 in South Australia at 2026 flat tariff rates. Adding active Jellyfin streaming should be calculated separately from measured transcode wattage, annual streaming hours and the household's actual tariff.
Does Docker itself use significant power on a NAS?
Docker runtime overhead is generally small compared with active workloads, but its wattage depends on the host and should be measured on the target NAS. The power comes from what the containers are doing. An idle container consumes almost no CPU and therefore almost no power. A container actively transcoding video or running machine learning inference can consume 20-80W depending on the workload.
Can I reduce NAS power consumption without removing containers?
Yes. The most effective options are: enabling hardware transcoding in Jellyfin (cuts transcode power from 40-80W to 8-15W if your NAS CPU supports it), adjusting container polling intervals to allow disk hibernation, and scheduling heavy tasks like Immich ML scans to off-peak hours. Choosing an ARM-based NAS over an Intel Core model at purchase time is the biggest single decision.
Which Australian states have the highest electricity costs for running a NAS?
South Australia consistently has the highest flat residential tariff, around $0.44-0.48/kWh in 2026, making it the most expensive state to run 24/7 hardware. The ACT and Victoria are at the lower end. For identical hardware and usage, the cost difference between South Australia and the ACT depends on the household's actual retailer plan; current flat standing-offer rates are not nearly two-to-one.
Use the NTKIT NAS Power Calculator to enter your actual state electricity rate, NAS wattage, and usage hours for a personalised annual cost figure.