I spent 500+/month on AWS. Four-month payback. The cloud is a tax on people who will not learn infrastructure.
The Economics
The math is simple. A mid-range homelab -- Dream Machine Pro Max, Layer 3 PoE switch, NVMe storage, compute nodes -- costs roughly what four months of equivalent cloud spend would. After that, you run at the cost of electricity: about $25/month in my case.
Cloud vendors know this. Their pricing models are optimized for teams that never do the calculation. They pay $500/month for workloads that need 16GB of RAM and 4 cores because standing up infrastructure feels harder than clicking "deploy."
It is not harder. It is different work. And the skills transfer directly to production.
What the Setup Looks Like
Six VLANs segment traffic: management, lab (Kubernetes, storage), IoT, media, guest, security cameras. Layer 3 switch handles inter-VLAN routing. 10G fiber connects core components. PoE powers cameras, sensors, and access points from a single switch.
Storage runs RAID 10 across 4x 16TB drives on a UNVR, with a dedicated NAS for backups. Allocation: 40% lab (VMs, containers, test data), 30% security footage at 30-day retention, 20% media, 10% system backups.
Services: Prometheus and Grafana for monitoring, IPS/IDS with custom rulesets, a Kubernetes cluster for orchestration, CI/CD pipelines for testing.
What Went Wrong
WiFi dead zones. Initial AP placement left gaps. Mesh networking with wireless uplink fixed it. Should have done a site survey first.
PoE budget overflow. Cameras, sensors, and APs drew more than initial calculations predicted. Upgraded the PSU and implemented power scheduling. Now running with 25% headroom.
Storage I/O contention. Continuous security recording saturated disk bandwidth during backup windows. Edge caching and storage tiering cut main storage I/O by 70%.
None of these problems were unique. They are the same problems you hit in a production data center. Solving them at home costs a weekend, not an incident page.
The Real Argument
The serverless advocates are right that infrastructure is work. They are wrong that avoiding it makes you more productive.
Every Lambda cold start you debug, every surprise cloud bill you explain, every vendor lock-in you navigate -- that is also work. Work where you learn nothing transferable and own nothing at the end.
Running a homelab teaches networking, storage architecture, monitoring, and capacity planning. These are the skills that separate senior infrastructure engineers from people who click buttons in a console. The money you save is secondary. The understanding you build is the actual return.