Price points
Choose a home lab by budget and workload.
A home lab does not need to be expensive to be useful. The best configuration depends on what you want to run, how much maintenance you are willing to do, and how long you want the setup to stay relevant.
Find a sensible starting tier
Answer three questions for a starting point—not a shopping mandate. The right build is the smallest one that meets your real needs.
A multi-node setup does not automatically provide high availability. Reliable failover takes compatible storage, networking, configuration, and tested recovery—and never replaces backups.
Home lab tiers at a glance
One quiet, low-power system for learning and running a useful set of home services.
- Single mini PC or refurbished desktop
- Proxmox VE or Ubuntu Server with Docker
- Basic network switch and backup storage
- Good fit for a first always-on server
A multi-node lab with dedicated shared storage and managed networking.
- Two or three small compute nodes
- NAS plus a separate backup plan
- Managed networking and VLANs
- Room to learn clustering and orchestration
Higher-performance, production-like infrastructure for demanding workloads and realistic failure testing.
- High-memory compute and optional GPU
- Fast storage and multi-gigabit networking
- Redundant services and expanded backups
- Plan carefully for power, heat, and noise
Example builds by tier
These are starting points, not fixed shopping lists. Prices change, and used hardware can offer very different value.
One quiet, low-power box that does a lot.
| Component | Example | Approx. cost |
|---|---|---|
| Compute | Used Dell OptiPlex, HP EliteDesk, or Lenovo ThinkCentre Tiny; 8th-gen-or-newer Core i5, 16 GB RAM | $120–$180 |
| Memory upgrade | Compatible 16 GB DDR4 SODIMM module to bring the system to 32 GB; verify the mini PC's memory type and slot configuration | $45–$75 |
| Boot and VM storage | 500 GB–1 TB NVMe SSD | $55–$95 |
| Bulk storage | 4 TB external USB HDD; a 4 TB SATA SSD is a different, higher-cost option | $100–$150 |
| Networking | 5-port gigabit switch, unmanaged or lightly managed | $15–$25 |
| Power protection | Small UPS, around 600 VA | $55–$70 |
Software: Proxmox VE or Ubuntu Server with Docker.
Good for
- Pi-hole or AdGuard Home for network-wide ad blocking
- Jellyfin or Plex for one or two simultaneous 1080p streams; Intel Quick Sync can help with transcoding
- Home Assistant for smart-home automation
- Nextcloud, Vaultwarden, or Paperless-ngx
- Uptime Kuma and Grafana for basic monitoring
- Laptop backups with Restic or Duplicati
- A few small VMs for learning Linux, scripting, and Ansible
A small cluster with shared storage, managed networking, and a path to redundancy.
| Component | Example | Approx. cost |
|---|---|---|
| Compute (3 nodes) | Three Lenovo M920q, Dell 3080 Micro, or Minisforum MS-01 systems; Core i5/i7 and 32 GB RAM each | $750–$1,050 |
| Node storage | Three 1 TB NVMe drives for local storage or replication; choose endurance and warranty to suit the workload | $150–$270 |
| NAS | Used Synology DS920+ or a DIY build with an Intel N100 board and four bays | $250–$400 |
| NAS drives | Four 8 TB NAS-rated HDDs at roughly $150–$220 each; usable capacity depends on RAID/SHR layout and formatting | $600–$880 |
| Switch | Managed 8-port 2.5 GbE switch | $80–$150 |
| Power protection | Around 1,000 VA UPS with USB monitoring | $100–$150 |
| Rack or shelf | Wall-mount 6U rack or suitable shelf | $40–$80 |
Software: Proxmox VE; add k3s or Docker Swarm for orchestration, and TrueNAS or the NAS vendor's OS for storage. Proxmox high availability requires appropriate cluster design and shared or replicated storage; a cluster alone does not guarantee automatic failover.
Good for
- Practice clustering and recovery; test VM migration and failover before relying on it
- Kubernetes practice with k3s, Argo CD, and GitOps workflows
- A media stack such as Jellyfin with media stored on the NAS
- Self-hosted Git, CI runners, and a private container registry
- VLANs for separating IoT, guest, and lab traffic
- A 3-2-1 backup plan using Proxmox Backup Server and an off-site copy
- Home Assistant or Frigate NVR; camera workloads may benefit from a Coral TPU
Enterprise-style infrastructure with high performance, redundant networking, and room for GPU or AI workloads.
| Component | Example | Approx. cost |
|---|---|---|
| Compute (3 nodes) | Used Dell R640/R740 or Supermicro rack servers with dual Xeon Silver CPUs and 128–256 GB RAM each | $1,500–$2,400 |
| Modern compute alternative | Three Ryzen 9 or EPYC small-form-factor systems with 64–128 GB RAM each, instead of the used rack servers above | $2,400–$3,500 |
| GPU (optional) | Used RTX 3090 (24 GB) or RTX 4070 Ti Super for AI or transcoding | $700–$1,000 |
| Fast storage | Four 2 TB enterprise NVMe drives for a Ceph or ZFS pool | $500–$900 |
| NAS and bulk storage | Eight-bay chassis plus eight 16 TB NAS-rated HDDs, roughly $200–$280 per drive; raw capacity is not the same as usable capacity | $2,000–$3,240 |
| Core switching | 10/25 GbE switch, such as a MikroTik CRS-series or Ubiquiti Aggregation switch | $250–$700 |
| Firewall/router | Protectli or Netgate appliance running pfSense or OPNsense | $200–$400 |
| Rack | Enclosed 12U–22U rack with PDUs | $200–$500 |
| Power protection | 1,500 VA online UPS and metered PDU | $300–$600 |
| Cooling and noise | Rack fans, sound management, and possible electrical work | $50–$150 |
Software: Proxmox VE with Ceph where the network and disks can support it; Kubernetes (Talos or RKE2), OPNsense, TrueNAS SCALE, and Prometheus/Grafana/Loki are possible parts of a larger stack.
Good for
- Local AI workloads with Ollama or vLLM, and image generation on a suitable GPU
- Production-like Kubernetes with multi-node storage, ingress, and observability
- Infrastructure as code with Terraform and Ansible to rebuild from a repository
- Personal cloud services such as Matrix, Immich, and Nextcloud
- Isolated security training networks for authorized penetration-testing labs and capture-the-flag practice
- 10 GbE or faster networking and redundant paths for controlled failure testing
- Active Directory and identity-management practice for certification preparation
Notes across all tiers
- Component prices are broad US retail estimates, not live quotes. Drive class, warranty, memory type, seller, and promotions can shift prices significantly; compare current listings and verify compatibility before buying.
- Power cost matters: a low-power starter system may idle around 10–20 W, while an advanced rack can idle around 300–600 W. At roughly $0.15/kWh, those ranges are about $13–$26 and $394–$788 per year respectively; local electricity rates change the estimate.
- Enterprise rack servers can be loud and produce substantial heat. Make sure you have an appropriate location and power capacity before purchasing.
- Each tier can build on the previous one. A starter mini PC can later become an extra node, test server, or backup utility host.
- Drive redundancy and cluster failover are not backups. Keep separate, tested copies of important data, including an off-site copy where practical.