The Real Cost of Self-Hosting: Time, Power, and What I'd Pay for SaaS

"How much does your homelab cost to run?" is the question I get most often. The honest answer is complicated, because the cost of self-hosting isn't just electricity and hardware. It's time. And time is the part most people don't calculate.

I run 14 services on a two-node Kubernetes cluster. Here's what it actually costs, compared against what I'd pay if I used SaaS alternatives for everything. No rounding in my favor. No pretending my time is free.

The Hardware

Total hardware: ~,050

The Dell laptop was free. It was going to e-waste. The NUC was bought specifically for the homelab. The NAS was a project I'd have built regardless for media storage. If I'm being honest about attribution, the homelab-specific hardware cost is really just the NUC: .

Amortized over 3 years (reasonable for this hardware), that's /month for the full stack or /month if I only count the NUC.

Power Consumption

Total average draw: ~73W

At the US average electricity rate of /bin/zsh.16/kWh:

73W × 24 hours × 365 days = 639.5 kWh/year
639.5 kWh × /bin/zsh.16 = .32/year
= .53/month

For Colorado (where I am), rates are closer to /bin/zsh.14/kWh, so ~.50/month for electricity.

The laptop and NUC are remarkably efficient. Modern mobile processors idle at barely more than a light bulb. The NAS is the biggest draw because spinning disks don't sleep when NFS clients are connected.

The SaaS Comparison

Here's what I'd pay to replace each self-hosted service with its closest SaaS equivalent:

Total SaaS cost: ~/month

Some of these comparisons are imperfect. Spotify isn't the same as owning your music library: you're renting access to their catalog. Audible gives you DRM-locked files. YNAB does things Actual Budget doesn't, and vice versa. But for the purpose of "what would I pay to get roughly equivalent functionality without self-hosting," /month is a reasonable estimate.

The Math

On pure dollars, self-hosting saves about /month or /year. After the hardware pays for itself (roughly 2.5 years at this savings rate), the ongoing cost is just electricity and the domain: about /month.

If I only count the NUC as homelab-specific hardware, the break-even drops to under a year.

The Part Nobody Calculates: Time

Here's where honesty gets uncomfortable.

Initial setup: Building the Kubernetes cluster from scratch (Talos Linux, Flux, Longhorn, Traefik, Authentik, every application) took roughly 80-100 hours over several weeks. That's not continuous work; it's evenings and weekends. But it's real time.

Ongoing maintenance: In a typical month, I spend 4-8 hours on homelab work. That includes updates (Renovate handles most of it, but I review and merge), debugging (something always breaks), adding new features, and writing about it.

Incident response: Maybe once every two months, something needs urgent attention. A Talos upgrade breaks the control plane. A PVC fills up. Cloudflare Tunnel drops. These take 30 minutes to 2 hours each.

Conservatively: 6 hours/month average on maintenance and ops.

If I value my time at my professional rate, the homelab costs far more than SaaS. If I value my time at /bin/zsh (hobby time I'd spend on something anyway), self-hosting is clearly cheaper. Reality is somewhere in between.

Where Self-Hosting Wins

Data ownership. My health data from Garmin is in my database, not just Garmin's. My music library is files I own, not a streaming license. My RSS reading history doesn't train someone else's recommendation algorithm. When a service shuts down (Google Reader, anyone?), I keep my data.

No monthly creep. SaaS costs increase. Every year, Spotify raises prices. YNAB went from /year to /year to /year. Self-hosting costs are front-loaded and decrease over time as hardware amortizes.

Customization. Life Hub doesn't exist as a SaaS product. The specific combination of data sources I pull together (Garmin + Navidrome + Miniflux + GitHub + finance) with AI-powered advisors and personalized dashboards isn't something you can buy. Self-hosting lets me build exactly what I want.

Skills development. Every problem I solve in the homelab (network policies, GitOps, secret management, DR testing, monitoring) maps directly to my professional work and certification tracks (CKA, CKAD, CISSP). The homelab is a production-grade learning environment. That has real career value that's hard to quantify.

Where Self-Hosting Loses

Reliability. Spotify has five nines of availability. My Navidrome has maybe three nines on a good month. When I'm upgrading Talos or migrating PVCs, my music is offline. My family notices. SaaS services have teams dedicated to keeping things running. I have me, and I sleep.

Mobile experience. SaaS apps have dedicated mobile teams building native experiences. My self-hosted apps work on mobile, but they're web apps or use third-party clients. The experience is functional, not polished.

Setup time. Signing up for YNAB takes 5 minutes. Setting up Actual Budget on Kubernetes takes an evening. For services where the value is in using them, not running them, SaaS wins on time-to-value.

Family buy-in. My wife doesn't care that Navidrome is running on Kubernetes with GitOps. She cares that the music plays when she presses play. Every minute of downtime erodes the family's tolerance for the homelab project. SaaS services don't require explaining why dinner music stopped because of a CSI driver update.

The Honest Verdict

Self-hosting is cheaper than SaaS if and only if you value your time at zero. Since this is hobby time I'd spend on technical projects anyway, I'm comfortable with that framing. The /month savings is a bonus. The real value is in data ownership, skills development, and building things that don't exist as products.

If I had to choose a subset of services to self-host and pay for the rest, I'd keep:

• Self-hosted: Navidrome (music ownership), Life Hub (doesn't exist as SaaS), Vaultwarden (trust), Miniflux (privacy)
• Would pay for: Ghost Pro (not worth running a CMS just for a blog), YNAB (better product than Actual), Audible (more content than my library)

But I don't have to choose, because I already built the cluster. The marginal cost of adding one more service to an existing Kubernetes cluster is nearly zero. That's the real economics of self-hosting: high fixed costs, near-zero marginal costs. Once you have the platform, everything else is basically free.

Is it rational? Probably not, by a strict dollars-per-hour analysis. Is it worth it? For me, absolutely. The homelab is where I learn, where I build, and where I keep the data that matters to me. You can't put a SaaS price tag on that.