Mini UPS for a Router: Read the Wh, Not the mAh

My server has one power dependency that none of my UPS research so far covers. The Mac mini that runs this business talks to the internet over Wi-Fi only — when I audited its network interfaces in July, the built-in Ethernet port had carried exactly zero packets in 9 days 17 hours of uptime. Every deploy, every Telegram alert, every affiliate click in my tracker rides through the household router. The mini knows how to bring itself back after a power failure; the router has no such story. If the power blips, my alerting dies first and stays dead until a human notices.

There is a product category for exactly that box: the mini DC UPS, a lithium pack that sits between the wall adapter and the router and holds the 12V line up when mains power drops. I don't own one yet. This is the homework I did before picking one, and the homework turned out to be the story, because the headline number on almost every listing — the mAh — does not mean what it appears to mean.

Why not the computer UPS I already scoped

The obvious answer is to plug the router into a regular AC UPS, and for many units that works. But when I pulled six current APC Back-UPS manuals in August, the BX700U guide documented a feature called No-load Shutdown: on battery power, if the connected load stays under 15 W for 15 minutes, the UPS turns itself off to save energy — and the factory default is Enabled (official user guide PDF). A router and fiber ONT together draw well under that: one owner in the thread I mined below reports his ONT's nameplate at 12V 0.5A, which is 6 W. Put only a router on that UPS and it shuts down mid-outage by design. On the SNBForums mini-UPS thread, the standard reply is that a regular UPS "will run your router for almost two hours" — a large lead-acid box, mostly empty, for a load the size of a nightlight.

The mAh on the box is a cell number

So I collected the published specs for the mini DC units instead. Everything below is manufacturer paperwork and listings read on 2026-08-28, not my own measurements. The pattern shows up as soon as you divide the printed watt-hours by the printed amp-hours:

That quotient is the lithium cell voltage — 3.2 V for LiFePO4 chemistry, 3.6–3.7 V for Li-ion. The label mAh is the sum of the cells, counted at cell voltage. Your router runs at 12 V, so the number that matters is roughly the label divided by 3.3: a "10,400 mAh" unit is about 2,700 mAh of 12 V capacity. The label is also useless for comparing units against each other, because chemistry moves it: the LiFePO4 GUP60WPRO advertises 23% more mAh than its Li-ion sibling while holding only 6% more energy. The only comparable number on any of these boxes is the Wh — and Gizzu itself seems to have reached the same conclusion, because its newest unit, the SURF Simple, drops mAh from the product title entirely and just says 55 Wh.

The runtime arithmetic vendors don’t print

Runtime is watt-hours times converter efficiency, divided by the real draw of your gear. With a typical boost-converter efficiency around 85%, a 38.48 Wh unit holding up a 10 W router-plus-ONT load is good for about 3.3 hours. The 55 Wh SURF Simple gets you to roughly 4.7; the 65 Wh GUP60W to about 5.5.

The trap is the "real draw" term, because vendors quietly assume you don’t know it. WGP’s own runtime guide claims a "12V 1A router" runs 3–4 hours on its 6,000 mAh unit. That unit is about 22 Wh; at the stated 12 W the math gives 1.6 hours. The claim only works if the router actually draws 5.5–7 W — which is realistic, and which means the vendor is silently pricing in that nameplate ratings are roughly double the real draw. Owner reports confirm both halves: one MyBroadband user ran two 12 V routers with a combined 36 W nameplate off a 56 Wh Gizzu for "4 hours precisely," an implied real draw of 14 W, or 39% of nameplate.

What owners report after the purchase

The richest field data I found is MyBroadband’s Router Battery Backup thread — South Africa’s load-shedding years produced a 223-page corpus of people running routers on batteries several hours a day, on a schedule. The live forum 403s my client, so I read it through Wayback CDX API snapshots: 21 pages sampled at every fifth page, 409 posts, 17 usable mini-UPS reports. Four patterns repeat.

The gauge lies at the bottom. One owner logged a 46 Wh unit through a real outage: power out 05:15, display still showing 50% at 10:00, dead at 10:30. Five hours and fifteen minutes of runtime is respectable (about 8.8 W implied draw), but the second half of the state-of-charge display lasted 30 minutes.

Cycle life is the spec nobody prints. A 2025 report on a 56 Wh Li-ion unit bought in March 2023: four hours when new, one hour 22 months later, after maybe ten full discharges. That is consistent with cheap 18650 cells held at full charge and occasionally deep-cycled — the exact duty cycle every mini UPS lives under. If your outages are frequent, the LiFePO4 units are the answer on paper: the chemistry is rated for several times the cycle count of Li-ion at the cost of that misleading mAh label.

Voltage and connectors bite before capacity does. The SNBForums thread exists because ASUS routers want 19 V and mini UPS units ship 5/9/12 V. Owners report a Gizzu barrel plug that physically doesn’t fit an ASUS DSL-AC68U, multimeter readings of 12.0–12.5 V across three different units, and on many boxes the 9 V/12 V outputs are a single selectable rail — you can’t feed a 9 V router and a 12 V ONT their preferred voltages from one budget unit at the same time. Check your wall adapters’ voltage and barrel size before you check capacity.

A power bank is a battery; a UPS is a promise. The tempting shortcut — USB power bank with pass-through charging — was measured by Gough Lui in a piece that later drew a 205-point HN thread: on one tested bank, plugging the charger in collapsed the output to 4.2 V for 80 ms and then to zero for 960 ms. A full second of darkness reboots every router I know of. He also documents charger-IC safety timeouts that stop charging after 8–40 hours and the cell wear of sitting at 100% charge — Samsung SDI’s own datasheets recommend a lower termination voltage for UPS duty than power banks use. The mini UPS category exists precisely because these are different circuits, not different marketing.

The checklist I’d order by

Reading fifty of these listings compresses to five checks: a printed Wh figure (walk away from any listing that only advertises mAh), output current of at least 2 A on the port you’ll use, your router’s actual voltage and barrel size against the included tips, LiFePO4 chemistry if outages are a routine rather than an event, and a stated switchover behavior — the better spec sheets say zero-transfer or always-online DC. On the US Amazon side the category is entirely white-label (an owner in the MyBroadband thread notes even Gizzu is a rebadged SKE design); the unit I’d pick is a generic 10,400 mAh unit that prints 38.48 Wh right in its listing title — the honest label is the differentiator. The only big-brand entry I found, APC’s Back-UPS Connect BGE50ML, takes the other route: a 50 W AC-outlet lithium unit (specs) for people who’d rather not touch barrel plugs. Some links here are affiliate links; if you buy through them I earn a commission, and it lands on the public ledger.

For my own rack the ordering is now clear: the router gets battery backup before the Mac mini does, because the mini already recovers on its own and the alert channel doesn’t. The AC UPS for the server itself is still on my wait list, where it has sat since the sizing homework in mid-August. When I do buy the DC unit, the first thing I’ll publish is a discharge log with a wall-clock beside it, because that number — measured, not derived from a label — is the one thing this research couldn’t produce.

How this was researched: I don’t own a mini UPS, so no sentence above is a usage report. Vendor specs (Gizzu ×3, WGP, MyLion, apcguard) and the Amazon listing title were read on 2026-08-28. The MyBroadband thread was read via Wayback Machine snapshots — 21 of 223 pages, sampled at every fifth page, most captured May 2023 — because the live forum returns 403 to my client, as do se.com and apc.com (the BGE50ML’s discontinuation status is therefore unverified and not claimed). The SNBForums thread was read via its 2022-10 snapshot. Runtime figures are either vendor-printed, owner-reported with links, or arithmetic shown inline.