Frank
My electric bill went up forty-three dollars the month I spun up my fourth NAS and added two more VMs to the homelab. I remember staring at that bill thinking, “I am literally paying a utility company to store my own stuff.” And then the obvious follow-up thought: “This is still cheaper than what AWS would charge me for the same capacity.”
That math matters. Hold onto it.
Right now there’s a legitimate trend happening in enterprise IT called cloud repatriation. Companies that spent the last decade migrating everything to AWS, Azure, or GCP are quietly moving workloads back on-premises. Not all of them. Not all the workloads. But more than the cloud vendors would like you to know about, and more than the trade press covered when the original migrations were the exciting story.
The thing is, homelab people already ran this experiment. We ran it broke, in spare bedrooms and garages, with used server hardware from eBay and a lot of late nights reading documentation. The enterprise world is just now catching up to conclusions we reached years ago, mostly because the bill finally got big enough that somebody in finance started asking questions.
I am not claiming homelab operators are smarter than enterprise architects. We are absolutely not. What we are is cheaper, which means we feel pain faster and adjust faster. There’s no budget approval process between me and the realization that a particular service costs more than it’s worth.
That feedback loop is the whole story.
What Homelab Actually Taught Me About Cloud Economics
When I started adding services to my homelab, cloud was the default recommendation everywhere. Spin up an EC2 instance. Use S3 for storage. Let AWS handle the infrastructure. The advice made sense on the surface, especially if you weren’t already sitting on hardware.
But I was sitting on hardware. And I kept running the numbers.
Storage is the obvious one. I’m currently managing close to 70TB across four NAS units at home. Rollo handles the bulk of it, movies and TV and music. The other three handle specialized workloads. The combined cost of that hardware, amortized over a reasonable lifespan, is a fraction of what equivalent cloud storage would run per month. Not over a decade. Per month.
Compute tells a similar story, though it’s messier. Some of my homelab VMs are running services that genuinely benefit from being local: low latency access, no egress fees, no dependency on an internet connection. My Emby server pulling a movie from Rollo across my local network is not the same experience as streaming from a cloud-hosted service, and not just for quality reasons. It’s mine. If Emby the company folds tomorrow, my media doesn’t go with it.
Other workloads, I’ve kept in the cloud or used cloud APIs specifically because local hosting would be absurd. I’m not spinning up GPU instances in my garage to run my own language model. I’m hitting the Claude API and paying per token like a reasonable person. That distinction matters: the argument was never “cloud bad, local good.” The argument is that the original calculus, “cloud is always simpler and usually cheaper,” turned out to be wrong at scale, and homelab people discovered that first because we were working at personal scale the whole time.
What I see now in the enterprise repatriation coverage is exactly this realization arriving with a seven-figure invoice attached. Companies that migrated their on-prem databases to cloud-managed services are discovering that egress costs alone can be brutal. Workloads that run 24/7 at predictable volume are almost always cheaper on owned hardware. The elasticity argument, “you only pay for what you use,” only holds up for genuinely variable workloads. A hospital system running the same reporting jobs every night doesn’t have variable workload. It has a utility bill that should be infrastructure.
I work in healthcare IT. I’ve watched this conversation happen from a different angle than most homelab people. The drivers in enterprise are compliance complexity, vendor lock-in anxiety, and egress costs that nobody fully modeled in the original business case. The drivers in my homelab were simpler: I could see exactly what I was spending, and I kept doing the math.
There’s a note in my notes app from sometime in 2022 that just says “cloud storage is rent, NAS is mortgage.” I have no memory of writing it, but it’s not wrong.
The broader lesson here isn’t really about cloud versus on-prem. It’s about who absorbs the feedback first. In enterprise IT, there are layers between the technical decision and the budget impact. Procurement cycles, multi-year contracts, organizational inertia. By the time a bad cloud cost decision surfaces as a real problem, years have passed and migrations have been completed and the person who originally signed the contract might not even work there anymore.
In a homelab, you get the bill next month. That compression of consequence is genuinely educational.
I’m not saying run your homelab like a business. That’s a miserable way to run a homelab, and I’ve tried it. What I am saying is that the homelab environment, resource-constrained and personally accountable, forces a kind of economic clarity that large organizations have to deliberately manufacture. The enterprise world is manufacturing it now, post-hoc, because the invoices finally made the conversation unavoidable.
The hardware sitting in my home office didn’t predict the future. It just couldn’t afford to get it wrong. That’s a different thing. But the conclusions landed in the same place.
The next wave of cloud spending won’t look like the last one. It’ll be more deliberate, more workload-specific, and a lot more skeptical of vendor promises about total cost of ownership. Every CFO who just approved a repatriation project is learning something homelab people learned by necessity, years earlier, for the cost of a used server and a slightly higher electric bill.
Forty-three dollars a month feels like a reasonable tuition rate, looking back.
