Low-Power Homelab: Run 15+ Services on Under 25 Watts
Build a homelab that idles under 25W while running 15+ services. Hardware picks, container density tips, power budget table, and honest trade-offs.
This post contains affiliate links. If you buy through them, I earn a small commission at no extra cost to you.
Most homelab builds drift toward more hardware over time. More nodes, a NAS, an old switch, a spare Pi running one service. Each addition makes sense in isolation. Together they quietly add up to a meaningful power bill and a more complicated setup than you actually need.
The other path is consolidation. One low-TDP machine, container density over VMs where it makes sense, SSD-only storage, nothing spinning. The result is a homelab that runs 15 or more services, stays under 25 watts at idle, and costs about $28 a year to keep powered at US average electricity rates.
This is the build guide for that setup. If you already have hardware running and want to measure what it actually costs you, the companion piece on measuring and reducing homelab power covers that side of the problem. This piece is about choosing hardware from the start so you’re building for efficiency, not retrofitting it.
The Core Principle: One Machine, High Density
The single biggest factor in a low-power homelab is how many devices are always on. Every device has a baseline idle draw that never goes to zero. A router, a NAS, a server, a switch: each one adds 10-30W that runs regardless of whether anything is actually happening.
Consolidating onto one machine eliminates that multiplier. If your server handles DNS filtering, reverse proxy, media indexing, password management, document storage, and monitoring all in one place, you’ve replaced four or five devices with one.
The modern N100 and N150 mini PCs are built for this. Intel’s N-series processors are efficiency-first designs. Not the stripped-down Celerons from five years ago, but genuinely capable chips with four cores, hardware video decode, and AES acceleration, all at 6-15W TDP. Under a normal mixed-service load, these machines idle between 8 and 15 watts. Heavy Jellyfin transcoding or a large backup job pushes them to 20-25W. That’s the ceiling, not the floor.
The Hardware
For a sub-25W homelab that can comfortably run 15+ services, you want:
Processor: Intel N100 or N150. Both are quad-core Alder Lake-N chips. The N100 maxes out at 6W TDP; the N150 is rated at 6W but sustains slightly higher clock speeds under load. Either works. Avoid Celeron N4xxx/N5xxx. Different architecture, slower, older node.
RAM: 16GB minimum, 32GB preferred. Most of these machines use LPDDR5, which is soldered. You get what comes in the box. The Beelink EQ12 Pro ships with 16GB. That’s enough for 15 lightweight services with headroom. If you plan to run heavier workloads alongside (Home Assistant + a few automation containers, a full Nextcloud instance, etc.), the extra overhead starts to matter.
Storage: NVMe only. No spinning drives attached to this machine. A 500GB NVMe handles the OS, Proxmox, and 15 LXC containers with room to spare if you’re not storing large media libraries locally. A 2TB NVMe like the Crucial P3 Plus in the secondary M.2 slot gives you local media or backup storage without adding any spinning rust.
Networking: 2.5GbE is standard on current-gen N100 mini PCs. This is fine. You don’t need 10GbE for a single-host homelab.
The recommended unit: The Beelink EQ12 Pro hits all of this. N100, 16GB LPDDR5, 500GB NVMe, 2.5GbE, fits in a hand, quiet under load. Around $165 new. If you want to spend less, the Beelink S12 Pro is the same platform at a lower price: 8GB RAM and 256GB storage, with standard SO-DIMM slots so you can upgrade RAM separately.
For a deeper look at new mini PC options across price tiers, the 2026 mini PC roundup breaks down the full field. If you want to stretch your budget further on used hardware, the corporate SFF guide covers Lenovo, HP, and Dell refurb picks.
The Services List
Here’s a realistic 15-service stack that runs comfortably on an N100 with 16GB RAM. These are all container-based and none of them are resource hogs in normal operation.
| Service | What It Does | Avg. RAM |
|---|---|---|
| Pi-hole or AdGuard Home | Network-level DNS ad blocking | 80-120MB |
| Nginx Proxy Manager | Reverse proxy + SSL termination | 80MB |
| Vaultwarden | Self-hosted Bitwarden password manager | 20-50MB |
| Uptime Kuma | Service monitoring and alerting | 80-150MB |
| Jellyfin | Media server (indexing; see trade-offs below) | 200-400MB |
| Immich | Photo backup and organization | 300-600MB |
| Paperless-ngx | Document scanning and search | 200-400MB |
| Mealie | Recipe manager and meal planning | 80-150MB |
| Heimdall or Homarr | Dashboard / homepage | 50-80MB |
| Portainer | Container management UI | 50-100MB |
| Watchtower | Automatic container updates | 30MB |
| Gitea | Self-hosted Git | 80-150MB |
| Nextcloud (lightweight) | File sync and sharing | 150-300MB |
| Grafana + Prometheus | Metrics and dashboards | 200-400MB |
| WireGuard / Tailscale | VPN access | 20-40MB |
Total rough RAM: 1.5-3GB under normal load. That leaves 13-14GB free for headroom, the Proxmox host OS, and any additional containers you add later.
The Power Budget
This is where the math gets interesting. Under a mixed load with all 15 services running, here is what the hardware actually draws:
| Component | Idle Draw | Notes |
|---|---|---|
| N100 mini PC (host) | 8-12W | SSD-only, no external drives |
| 8-port managed switch | 4-7W | TP-Link TL-SG108E or equivalent |
| Router | 10-15W | Depends on your hardware |
| Server + switch total | 12-19W | Without router |
| Full setup total | 22-34W | Including router |
The mini PC handling all 15 services sits solidly under 15W idle. Add the switch and you’re still under 25W for the homelab hardware. Whether the router counts depends on your setup; most people already had a router running before they started a homelab, so the marginal cost is just the server and switch.
At 20W average for the server and switch combined, the annual electricity cost at $0.13/kWh works out to about $22.80 per year. Compared to the full cost picture of running a homelab, that’s the cheapest line item in the budget. It is genuinely nothing.
Proxmox + LXC: Container Density Done Right
Running 15 services on one machine requires thinking about density. The practical split for a low-power homelab:
Use LXC containers for stateless and standard Linux services. Pi-hole, Nginx Proxy Manager, Vaultwarden, Uptime Kuma, Gitea, and WireGuard all run cleanly in unprivileged LXC containers with minimal overhead. An LXC container starts in under a second and uses almost no memory beyond what the service itself needs.
Use Docker (inside one LXC, or directly on the host) for services that ship as Docker images. Immich, Paperless, Mealie, and Jellyfin all have well-maintained Docker Compose stacks. Run them from a single Docker LXC rather than from separate containers for each service.
Avoid VMs for lightweight services. A VM for Pi-hole wastes 512MB RAM on the guest OS before the service even starts. LXC gets you near-native performance with a fraction of the overhead. VMs are worth it when you need full kernel isolation or a non-Linux OS. Not for a standard self-hosted stack.
This architecture (Proxmox as the host, a handful of LXC containers, one or two Docker hosts) is what lets 16GB of RAM handle 15 services without memory pressure.
The Honest Trade-offs
Nothing about this setup is magic. Here is what you are giving up.
Transcoding is the big one. Jellyfin on an N100 will handle direct-play streaming of any format your TV supports natively. Hardware-accelerated transcoding via Quick Sync works and handles a couple of simultaneous streams. What it will not do is transcode 4K HEVC to 1080p for five simultaneous users. If your media server use case requires heavy simultaneous transcoding, this hardware is not the ceiling you want.
Storage ceiling is real. One 2TB NVMe is about 1.8TB usable. That handles a document library, a photo archive, and a modest media collection. It does not handle a 50TB media library. The minute you start adding USB drives or a NAS to this setup, the power advantage shrinks. If storage is the primary use case, start with a NAS-focused build instead.
No redundancy. This is a single point of failure. One machine, one drive. If it dies, everything is down until you replace it. For a home lab running personal services, that is usually an acceptable risk. If you are running anything business-critical or anything other people depend on, add a backup strategy at minimum, and consider a second node.
RAM is not expandable. Most N100 mini PCs use soldered LPDDR5. What you buy is what you have. If you need 32GB later, you are buying new hardware, not upgrading. The Beelink S12 Pro uses socketed DDR4, which is the exception. That one you can upgrade.
The Full Build Cost
| Item | Cost |
|---|---|
| Beelink EQ12 Pro (N100, 16GB, 500GB) | ~$165 |
| Optional: 2TB NVMe (Crucial P3 Plus) | ~$80 |
| TP-Link TL-SG108E 8-port switch | ~$35 |
| Total (without optional SSD) | ~$200 |
| Total (with 2TB storage) | ~$280 |
The core build comes in right at the $200 mark. For the full ongoing cost picture including what maintenance, drives, and incidentals look like over a year, the homelab cost breakdown has the math.
Is This Enough?
For most home users: yes.
If your services list is Pi-hole, a password manager, a VPN, self-hosted photos, a document scanner, a media server for existing rips, and a few utilities, an N100 mini PC handles all of it without getting warm. The 25W target is not a stretch goal. It is the realistic idle for this hardware class doing real work.
The case for going bigger is storage (if you have more media than an NVMe holds), redundancy (if you need HA), or compute (if you are running local AI inference or heavy transcoding). For those cases, the sub-25W constraint is probably the wrong starting point.
For everyone else, the single-machine, SSD-only, container-dense build is the one that makes sense both technically and financially. Low electric bill, no noise, fits anywhere, handles the services most home users actually run.
That’s worth more than a rack full of gear.