The Real Cost of a Home NAS Is Your Time

A home NAS beats a cloud subscription on the invoice and loses on the timesheet, and most comparisons only count one of them. The hardware bill is easy: a box, some drives, a few watts. The other bill is updates, monitoring, rebuilds, failed-drive RMAs and the evening you spend restoring from a backup to prove it works. Both are real, and the decision flips depending on which one you refuse to write down.

The Bill You Can See

Start with capacity you can actually use, not capacity you bought. Two drives in a mirror give you half of the raw figure. Four drives in a parity array give you three quarters, and parity is not a backup, it is a rebuild path: it survives one drive dying, and it survives nothing else. Deletion, ransomware, a bad sync and a controller writing garbage all propagate straight through it. Every storage argument that begins with "RAID" and ends with "backup" has skipped the only step that matters.

Then the drives themselves, which are consumables with a distribution rather than a lifetime. Backblaze publishes annualised failure rates from a fleet of hundreds of thousands of disks and reports that most drives run four or five years or more with low failure rates, which is the useful shape of the answer: a long low-risk middle and a tail where things fail in clusters. Plan for a replacement every few years, not a replacement every decade, and keep a cold spare because the failure will not be convenient.

Under the drives sits an arithmetic problem most buyers never run. Desktop drive datasheets quote a nonrecoverable read error rate of one in 10¹⁴ bits, and enterprise models one in 10¹⁵. Ten to the fourteenth bits is about 12.5 TB. Read a full multi-terabyte array during a rebuild and you are, statistically, asking for the one sector that cannot be corrected, which is why large parity arrays rebuild slowly into a smaller array and why a mirror is the honest choice for four drives in a house. Power is the smaller line item: forty watts continuous is about 350 kWh a year, which at fifteen cents a unit is roughly fifty dollars, plus the cost of not letting the drives cook in a closed cupboard.

The Bill You Can't See

Labour is the line item that decides these arguments and the one that never appears. A NAS is a small Linux server with a web interface, and it needs patching, monitoring, capacity planning, and someone who notices when a SMART attribute starts drifting. Pick your own hourly rate and be honest about the hours: an afternoon to build it, an hour a month to keep it healthy, and then a bad week when a drive fails and the rebuild takes two days during which you do not touch the array. Multiply that across five years and the number frequently exceeds the hardware it describes.

Then there is the offsite problem, which is where the self-hosting argument usually collapses. The 3-2-1 rule — three copies, two media, one offsite — cannot be satisfied by a box in your house no matter how many drives are in it. Fire, theft, flood and a power supply that fails shorted all take the array with them. So the honest architecture is not NAS versus cloud. It is NAS for the working copy and fast local reads, plus a cloud or offsite target for the third copy. The cloud does not disappear from the budget; it gets smaller and it stops being the thing you read from.

There is also an attack surface. A NAS holding every photo you own, reachable from the internet, is a target with a web login and a firmware update cadence set by a vendor who may lose interest. The defensible configuration is no port forwards, no exposed admin page, and remote access through a private overlay network or a VPN that terminates somewhere you control. That is another thing to configure, and another thing to get wrong in a way you will not notice for a year.

Where Cloud Genuinely Wins

For a modest dataset, cloud is simply better and there is no interesting argument. You are buying durable storage operated by people whose full-time job is durability, with versioning and geographic redundancy you would not build yourself, and with no hardware to RMA. The pricing structure matters more than the headline rate: Backblaze's B2, for example, includes free egress up to three times your average stored data and charges a cent per gigabyte beyond that, which is a reasonable shape for backup workloads and a punishing one for a media server that streams to friends every night. Read the egress terms before you pick a provider, not after.

Cloud loses where the workload is bulk reads of a library you already own. A subscription is a permanent cost that never depreciates into ownership, and it scales linearly with capacity forever, while a drive is a capital cost that then ages out. If you are storing tens of terabytes and reading them constantly, the crossover arrives quickly and the NAS wins on money alone. If you are storing two terabytes of photos and reading them occasionally, you are paying for the privilege of running a server, and the server is not the point.

What I Think

Buy the NAS for what cloud is bad at, not to avoid a monthly bill. Cloud is bad at large sequential reads of a library you already own, and it is bad at the restore that has to finish tonight. Everything else it does well enough that owning hardware to duplicate it is a hobby with a spreadsheet attached. If the spreadsheet is the reason, be honest that the spreadsheet excludes your evenings.

And if you do build one, budget for the third copy in the same breath as the first two. A NAS with no offsite copy is a single point of failure wearing a redundancy badge, and the failure mode is not a dead drive, it is the afternoon you discover the array was fine and the files were not.