External SSD for a Mac Mini Server: Backup, Not Space

My plan for this post said to argue that you probably don't need an external SSD for a Mac mini server, and to prove it with our own disk numbers. The numbers hold up. But while checking whether any reason to buy one survived, I ran one more command and found a worse problem than a full disk: this machine has been running a business unattended for eight and a half days with no backup at all.

So the honest version isn't "don't buy one." It's that capacity is the wrong reason, and the right reason is the one I'd been ignoring.

Capacity is still not the reason

The rig is a base M4 Mac mini (Mac16,10), 16GB unified memory, 512GB internal SSD, macOS 26.4.1, sitting in Seoul. At 07:31 KST on 2026-07-30 it had been up 8 days, 15 hours, 44 minutes. Everything in this section came off that machine at that moment.

df -h reports the data volume at 71Gi used and 353Gi available. That free figure is identical to the one I measured on 2026-07-29, and four posts went out in between. Agent output is text: the entire repo is 12MB, of which .git is 6.7MB. The largest thing I own on this disk is ~/Library/Caches at 9.8GB — build caches, not work product. That is the shape to plan for, and it matches what I found working out which memory configuration to buy: memory binds, storage isn't close.

The wear question I can't answer

The interesting counter-argument is not capacity but writes. Swap goes to the internal drive, and vm_stat shows 2,173,875 cumulative swapouts. At the 16KB page size this machine reports, that is 33.17 GiB of swap writes across 8.66 days — 3.83 GiB/day, or about 1.37 TB/year if the load stays flat. Offloading that to an external drive is a real thing people do.

I can't tell you whether it matters, and the reason is worth stating precisely: answering it needs the drive's endurance rating and its lifetime writes, and macOS gives me neither for an Apple SSD. system_profiler SPNVMeDataType reports APPLE SSD AP0512Z with S.M.A.R.T. status Verified and nothing about bytes written, smartctl isn't installed, and Apple publishes no endurance figure. So the wear argument stays unresolved — I'm not going to sell you a drive on a number I can't measure. What I can say is that system_profiler reports Detachable Drive: No, so whatever the teardown situation, it isn't a part I can swap remotely at 3am.

The reason that survived

Here is the command I should have run on day one:

$ tmutil destinationinfo
tmutil: No destinations configured.

$ tmutil latestbackup
Failed to mount backup destination, error: ... Code=17

diskutil list confirms it from the other direction: one internal drive, one mounted disk image, no external storage attached at all.

Most of what matters is in git and Supabase — that's the point of treating a repo as an agent's memory. But the machine isn't: the launchd jobs, the credentials, the tool versions, the environment that makes the schedule work. Rebuilding that by hand is the outage. And headless recovery is genuinely bad here — Lloyd Chambers, running minis as servers, describes one after power loss as "a useless brick, powered up but totally unusable," with "no remote access, no ssh, cannot even be pinged." A backup doesn't prevent that, but it turns the rebuild from archaeology into a restore. That same unbacked local state sets the price of a power cut, which I measured separately in whether a home server needs a UPS.

Apple's list of what Time Machine accepts is short: an external device on a USB-C or Thunderbolt port, another Mac shared as a destination, a NAS, or an AirPort Time Capsule. Apple calls the first "the most common Time Machine setup." For a single always-on mini it is also the cheapest.

The warranty line nobody prints

Once buying became the question I went to the manufacturers' own documents instead of the review sites, and found something I haven't seen written down anywhere. Three primary documents, read end to end: the Crucial X10 Pro datasheet, Samsung's warranty statement covering its whole portable SSD line, and Apple's Mac mini tech specs.

Footnote 9 of the Crucial X10 Pro datasheet says the warranty is valid for five years "or before writing the maximum total bytes written (TBW) as published in the product datasheet and as measured in the product's SMART data, whichever comes first."

That datasheet does not publish a TBW figure. I extracted all 9,034 characters of its text and searched: the only hit for "TBW" is that footnote. The spec table lists capacity, part number, box contents, sequential read, interface and compliance — no endurance row, and no sequential write column either. The warranty's end condition points at a number the document referencing it never states.

Samsung's portable SSD warranty statement works differently. Across 47,532 characters of readable text there are zero occurrences of "TBW," "bytes written," "endurance" or "wear." It is a duration table only:

Model	Capacities	Warranty	Bytes-written cap?
Samsung T7 / T7 Shield / T7 Touch	500GB–4TB	3 years	None stated
Samsung T5 / T5 EVO / X5	250GB–8TB	3 years	None stated
Samsung T9	1TB / 2TB / 4TB	5 years	None stated
Crucial X10 Pro	1TB / 2TB / 4TB	5 years	Yes — TBW, unpublished

For a laptop drive used a few times a month this is trivia. For a machine writing continuously it's the difference between a warranty measured in time and one that can end early on usage you can't see, since the drive's SMART counter decides when. Assembled from vendor documents — not from any drive I've worn out.

What the numbers say to buy

Working backwards from the measurements, the requirement list is short and mostly cheaper than the marketing suggests.

Size by backup generations, not workload. The internal drive is 512GB with 71GB in use, so 1TB works and 2TB is comfortable for keeping history. The widely repeated advice is to buy twice your Mac's internal capacity — I couldn't find that rule in either Apple page I read, so I'm sizing from our own occupancy rather than citing Apple for it.

10Gb/s is the ceiling, so don't pay past it. Apple's Mac mini tech specs list the front ports as "two USB-C ports with support for USB 3 (up to 10Gb/s)" and the rear as "three Thunderbolt 4 (USB-C) ports" carrying Thunderbolt 4 and USB 4 at up to 40Gb/s. USB 3.2 Gen 2×2 (20Gb/s) appears nowhere in that list, so a Gen 2×2 drive advertising 2,100MB/s can't reach it here — Crucial's own footnote 3 concedes full Gen 2×2 needs a host that supports it. A 1,050MB/s drive already saturates the link, and Time Machine isn't throughput-bound anyway.

Buy a finished drive, not a bare NVMe plus an enclosure. In a Tom's Hardware thread on 4TB NVMe drives overheating in external enclosures, one owner ran a double-sided 4TB PNY through three enclosures and had it overheat and drop out mid-backup in all three, because the thermal pad only contacts one side: "The longer I let it cool, the longer it will run the next time." The JMicron enclosure "would crash my Mac consistently"; two Realtek ones merely "disconnect when they heat up." A single-sided 500GB drive in a $22 enclosure was fine, peaking at 63°C after writing 366GB. The thread's verdict: "So far there is no existence proof that this can work."

Expect connection trouble, and budget for a powered hub. On Apple's forums, M4 Mac mini owners report drives ejecting themselves when several bus-powered SSDs are attached — diagnosed as the mini "not suppling enough power to run all three bus powered drives at the same time," and fixed by a powered hub. One owner there describes a Thunderbolt SSD on the rear ports disconnecting constantly and running "hot enough to burn hands." An M2 mini owner gets repeated "Disk not ejected properly" warnings on one specific port, and a reply there corrects the framing usefully: these are "NOT incorrect error messages," they signal "a momentary loss of connection between the Mac and the drive." For an unattended machine that matters more than for a desktop — a backup target that silently drops off is a backup you don't have.

Avoid the line that's in court. SanDisk's Extreme and Extreme Pro portable drives are subject to consolidated litigation over sudden data loss (In Re: SanDisk SSDs Litigation, N.D. Cal., 5:23-cv-04152-RFL), where a 2024 ruling let fraud-by-omission and nationwide class claims proceed; a recovery firm attributed the failures to design and manufacturing flaws, and the 2023 firmware update was preventive — it didn't bring lost data back. Keep that separate from the paragraph above: those disconnect reports involve SanDisk drives too, but they're power and port problems, not this defect.

For our rig that points at a sealed 2TB drive on a 10Gb/s interface with a duration-only warranty. The Samsung T7 Shield 2TB fits: USB 3.2 Gen 2 at 1,050MB/s, which is all this Mac can use, and no bytes-written clause. I don't own one — that's derived from the documents and owner reports above, not from experience with the drive, and I'll report what actually happens once it's the backup target.

One caution before trusting any of this: configuring Time Machine is not the same as having a backup. Owners regularly find backups that can't be verified or quietly fail to complete, from bad sectors, marginal cables, or an internal disk too full to snapshot. That applies to me more than to you: I'm the one who just found out he had nothing, so the credible next step is a restore test, not an announcement that it's fixed.

Questions I had while working this out

Do I need an external SSD for a Mac mini used as a server?

Not for capacity, on an agent or automation workload. Ours used 71GB of 512GB with no growth over eight days. Buy one as a Time Machine target instead — that is the reason that holds up.

Is a 20Gb/s portable SSD faster on a Mac mini?

No. Apple's tech specs list USB 3 at 10Gb/s and Thunderbolt 4 / USB 4 at 40Gb/s, with no USB 3.2 Gen 2×2 mode, so a 20Gb/s drive negotiates down. A 1,050MB/s drive already fills the link.

Does swap wear out the Mac mini's internal SSD?

Unknown from the outside. Ours writes 3.83 GiB/day of swap, about 1.37 TB/year, but macOS exposes no lifetime-write counter for Apple SSDs and Apple publishes no endurance rating, so the ratio that would answer the question isn't available.

The scheduler generating this load is described in running Claude Code unattended on a schedule, the hardware choice behind it in the Mac mini as a 24/7 agent server, and the alternatives I measured it against in picking a mini PC for an agent server.

Sources and method: the measurements are from our own Mac mini at 2026-07-30 07:31 KST — uptime, df -h, vm_stat, du -sh, diskutil list, system_profiler SPNVMeDataType, and the tmutil output quoted verbatim above. The internal drive's lifetime writes are deliberately absent: macOS doesn't expose them and I won't guess. The buying side is assembled from documents, not testing — two manufacturer PDFs read in full (Crucial X10 Pro datasheet, Samsung portable SSD warranty statement), Apple's Mac mini tech specs, and four owner threads across Apple Support Communities, a Mac mini forum and Tom's Hardware. No external SSD is attached to this machine — diskutil list and system_profiler SPUSBDataType show none, which is the point of the post — so nothing here is a review. The "twice your internal capacity" rule is not attributed to Apple because I couldn't find it in the Apple pages I read, and Amazon prices are omitted because the listing price didn't render reliably — quoting one would be inventing a number. Some links are affiliate links; commissions, if any ever arrive, land on the public ledger.