35% of your hard drives are dead by year six [1]. That is Backblaze’s own finding from the fleet it runs, and it should decide your next homelab purchase. A motherboard has no moving part. A disk does.

So the upgrade that actually pays is younger disks on a calendar, not on a warning.

The big picture:

The advice going around hobby circles is to buy older compute and skip the new generation. That works. A board has no moving part, so a five-year-old one is a fine place to save money.

Storage is the exception. Backblaze’s survival study, built from its own storage servers, puts about 90% of drives alive at four years and 65% at six [1]. Life expectancy drops 2% to 2.5% a year for the first four years, then the fall speeds up [1].

The number most people plan around is the fleet average. In the second quarter of 2026 that was a 1.41% lifetime failure rate across 354,415 drives [2]. The figure is real, and it is the wrong one. Young 20-terabyte (TB) and larger drives pull it down. In the same table, models averaging five to eight years of age run 3.2% to 9.3% [2].

What strikes me is how quietly that gap sits in a table most owners never open. The average gets quoted. The age groups decide whether an array survives the year.

By the numbers

  • 65% — six-year survivors: Share of Backblaze’s own drives still running after six years, against 90% at four years, in a fleet the company tracks every day [1].
  • 1.41% — fleet lifetime rate: Annualized failure rate across 354,415 drives in Q2 2026, the average that hides the older groups [2].
  • 7.63% — aged 12TB model: Rate for one 12TB model at an average age of 83.8 months, close to five times the fleet lifetime figure and above the report’s 6.95% outlier line [2].
  • $24.72 a year — box running cost: A two-bay Network-Attached Storage (NAS) box at 16.31 watts in access mode, the maker’s own spec [4], priced at the 2025 US residential average from the EIA (U.S. Energy Information Administration) of 17.30 cents per kilowatt hour [3].

What I’d watch:

I keep coming back to the arithmetic, because a drive never fails at a good time. The useful question is what owners of four- and five-year-old arrays do now.

  • The calendar, not the alert: The people I know who have not lost data replace disks on a schedule, one at a time, before the array is degraded. The curve is the argument. After year four, the fall speeds up [1].
  • The array sum: If one disk has a 12.5% chance of dying in a year, four disks have about a 41% chance that at least one dies. That is 1 − (1 − 0.125)^4. At the fleet’s 1.41% average the same sum gives about 5.5%. Both figures are the desk’s arithmetic on Backblaze’s numbers, not a published measure, and the gap between them is the story.
  • The rebuild read: A rebuild reads every surviving disk end to end. The disks doing that reading are the same ones the table shows failing faster [2]. That is an inference about load, not a measured rebuild rate. Nobody publishes one for a four-bay home array.
  • The per-TB year: Price divided by capacity and expected years is the only honest carrying cost. An 8TB disk kept six years costs its price divided by 48 TB-years. That is the figure a refresh has to beat.
  • The cloud yardstick: Backblaze’s own object storage bills $6.95 per TB per 30 days at its 2026-10-17 list rate [5]. For 8TB that is $667.20 a year. Local storage still wins on price by a wide margin. What staying local costs is attention and risk, not dollars.

The catch

My read is that treating storage like compute is the trap. The difference is the part that spins.

The fleet is not a home NAS. Backblaze runs 200,000-plus drives in storage servers, turning all day, mostly between 4TB and 18TB [1]. A four-bay box in a spare room sees less use, which may mean fewer hours on the bearings. It also sits idle for long stretches, and idle is not the same as safe. I would not read the fleet numbers as a promise about a light-duty array either way.

Model matters as much as age. Backblaze notes that two of this quarter’s three outliers also had small populations, where a handful of failures moves the rate a long way [2]. The 7.63% figure is a real measurement and a noisy one.

The power case is weaker than it feels. At 16.31 watts and 17.30 cents a unit [3][4], the box costs about $25 a year to run. So “it is already paid for” is nearly true on electricity. The reason to replace disks is not the bill. It is the hours after a failure. Our own field note from a self-hosted mail weekend still stings: 26 hours of debugging to avoid a $72-a-year mailbox.

There is a real limit here too. I can give you the survival curve and the aged-model rates. I cannot give you the odds that a rebuild finishes cleanly on a five-year-old four-bay array. No one publishes that for home hardware.

At a glance

  • The Big Shift: A hard drive is a wear item with a published survival curve, 65% of Backblaze’s fleet alive at six years. The boards and chips people usually upgrade have no wearing part at all.
  • Why It Matters: The 1.41% fleet failure rate is an average over young drives. Aged models in the same report run 3.2% to 9.3%, and that gap is the difference between a quiet array and a degraded one.
  • What I’d Watch:
  • The replacement calendar: Whether owners of four- and five-year-old arrays replace disks one at a time on a schedule, instead of waiting for an alert.
  • The per-TB-year cost: What a replacement costs per TB-year at the capacity and service life the owner actually expects.
  • The rebuild window: How much of the array’s data is exposed while a new disk is filled from the survivors.
  • The Catch: The fleet is Backblaze’s, not a home NAS. Model matters as much as age, and no one publishes a rebuild failure rate for four-bay home arrays.

Sources

[1] Backblaze, “How Long Do Disk Drives Last?” — https://www.backblaze.com/blog/how-long-do-disk-drives-last [2] Backblaze, “Q2 2026 Drive Stats: Hard Drive Failure Rates” — https://www.backblaze.com/blog/backblaze-drive-stats-for-q2-2026 [3] U.S. Energy Information Administration, “Electricity explained: Prices and factors affecting prices” — https://www.eia.gov/energyexplained/electricity/prices-and-factors-affecting-prices.php [4] Synology, “DS223j” product specifications — https://www.synology.com/en-us/products/DS223j [5] Backblaze, “B2 Cloud Storage Pricing” — https://www.backblaze.com/cloud-storage/pricing