PoE Splitter USB-C: 20 Spec Sheets, 2 Reach 5V/5A

A PoE splitter with a USB-C plug is the cheapest way to put a Raspberry Pi 5 anywhere a network cable already reaches: power and data arrive on one wire, and the box hands you 5 V on Type-C plus an RJ45. The listings all lead with a current figure — 2.4 A, 3 A, 4 A, 5 A, one of them 6 A. I pulled the published specifications for 20 USB-C splitters and put them on the same axes: claimed standard, amps at 5 V, and Ethernet port speed. Two of the 20 claim the 5 A that Raspberry Pi asks for on a Pi 5. Four of the 20 have 10/100 ports. And when I divided the labels by what the PoE standards actually deliver at the far end of the cable, the two biggest numbers stopped making sense.

I own none of these splitters. My rig is a Mac mini with no PoE anywhere in it, so every product number below is either a line from a vendor document or a named person in a public thread, and I say which one each time. Some links are affiliate links, and commissions land on the public ledger. As an Amazon Associate I earn from qualifying purchases.

Start at the other end of the cable

A splitter cannot output more than the standard hands it, minus conversion loss. Those two ceilings are published. Texas Instruments application note SLVA221 works the 802.3af case out from the standard's own worst case — 44 V minimum source voltage, 100 m of cable, 20 Ω feed resistance — and lands on 12.95 W at the powered device's input. For 802.3at, EDN's PoE basics and beyond gives the familiar pair: 15.4 W at the port, 25.5 W at the device.

On the other side sits the load. Raspberry Pi's own getting started documentation lists the Pi 5 as 5 V at 5 A (27 W), and notes that a 5 V 3 A supply limits peripherals to 600 mA. That is the whole problem in one line: the board asks for 25 W on the 5 V rail, and 802.3at delivers 25.5 W to the splitter before the splitter converts anything.

The 20 splitters, on the same axes

Fifteen of them come from vendor specification pages (Waveshare, The Pi Hut, Adafruit, UCTRONICS, RAKwireless, 52Pi, POE Texas, Linovision, PROCET); the other five are retail listings, which is the only place their makers publish a number. All read on 2026-09-18. The Pi Hut's industrial 2.5 A unit is a Waveshare board, so I counted it once; the PROCET row covers three models with identical 5 V behaviour, which is how 18 rows hold 20 splitters.

Splitter	Claimed standard	5 V output	Ethernet
Waveshare Industrial Gigabit, 5 V 5 A	af/at	5 A (MAX)	10/100/1000
RAKwireless RAK9168	af/at	6 A (30 W)	10/100/1000
UCTRONICS U5259	at (af too)	4 A at · 2.5 A af	10/100/1000
52Pi Gigabit USB-C	af/at	4 A at · 2.5 A af	Gigabit
DSLRKIT 5 V 4 A Gigabit	at	4 A	Gigabit
Gigabit splitter, 5 V 3.5 A (17.5 W)	af/at	3.5 A	10/100/1000
UCTRONICS U6271	af/at	3 A	10/100/1000
POE Texas GAT-USBC	at	3 A at 5 V (USB PD)	10/100/1000
PROCET PT-PTC-AT / PT-PTC-G-AT / PT-PGC-WAT	at	3 A at 5 V (USB PD)	10/100/1000
DSLRKIT 5 V 3 A Gigabit	at	3 A	Gigabit
REVODATA 2.5G Type-C	af/at	3 A	2.5 Gbps
Waveshare POE-SPLITTER-TYPE-C	af/at	2.5 A (MAX)	10/100/1000
The Pi Hut USB-C Gigabit, 5 V 2.4 A	af/at	2.4 A	10/100/1000
UCTRONICS U6115 (UC-3AF-USBC)	af	2.4 A	10/100/1000
ANVISION 5 V 2.4 A	af	2.4 A	10/100
Linovision POE-SP01C	af	2.4 A	10/100
Adafruit 4552	af	2 A	100 Mb
POE Texas AF-USBC	af	2 A	10/100

Counted up: two models claim 5 A or more, five claim 4 A or more, eight sit at 3–3.5 A, and seven top out at 2.5 A. Five are 802.3af only. Fifteen print both af and at on the box, and exactly two of those fifteen — the UCTRONICS U5259 and the 52Pi — tell you that the output changes with the source, publishing 4 A on 802.3at and 2.5 A on 802.3af. The other thirteen print the af and at logos side by side and then show you one number, the big one.

Divide the label by the ceiling

This is the part no listing does for you. Take the claimed output in watts, divide by what the standard puts at the device, and you get the conversion efficiency the label needs in order to be a continuous rating.

Label	Watts out	Standard's watts in	Efficiency the label needs
5 V 6 A	30 W	25.5 W (at)	118%
5 V 5 A	25 W	25.5 W (at)	98.0%
5 V 4 A	20 W	25.5 W (at)	78.4%
5 V 2.5 A	12.5 W	12.95 W (af)	96.5%
5 V 2.4 A	12 W	12.95 W (af)	92.7%
5 V 2 A	10 W	12.95 W (af)	77.2%

I assumed no efficiency figure to build that table; it is two published numbers divided. A 6 A claim on an 802.3af/at box is above the entire 30 W the PSE puts on the wire, never mind what reaches the far end. A 5 A claim needs an isolated step-down converter to lose essentially nothing. Waveshare's own wording is the tell — the spec line reads 5V 5A (MAX), and MAX is doing a lot of work there. On the 802.3af side the same arithmetic quietly indicts the common 2.4 A and 2.5 A labels: they need 93–97% efficiency to hold. The numbers that leave room are 4 A on 802.3at and 2 A on 802.3af.

The USB-PD splitters advertise a number you will not get at 5 V

Four of the twenty negotiate USB Power Delivery instead of hard-wiring 5 V, and their headline wattage describes the 20 V profile. POE Texas publishes the whole ladder for the GAT-USBC: 3A/5v, 2.4A/9v, 2A/12v, 1.7A/15v, 1.3A/20v, 25.5 W maximum. PROCET's at-class USB-C units follow the same shape — 5 V/3 A, 9 V/3 A, 12 V/3 A, 15 V/3 A, 20 V/2.25 A — under a 25 W to 27 W headline.

Someone hit this in public. In the Raspberry Pi forum thread Pi5 w/NVME - Need POE splitter to supply Req 27w, a user needed 27–30 W for a Pi 5 with NVMe off a switch that could supply it. Paul Hutch reported back on 2024-06-03 that the PROCET unit he found was “60W but only provides 15W @ 5V”, and suggested taking 12 V off a splitter and stepping it down to 5.1 V instead. A 60 W splitter that gives a Pi 15 W is not a defect; it is the PD profile table, printed nowhere near the headline.

Four of the twenty are 10/100, and one vendor's blog argues with Raspberry Pi

The Adafruit 4552, the POE Texas AF-USBC, the Linovision POE-SP01C and the ANVISION unit are all 10/100 only. That is the same trap I found on the source side of the link, where nine of 30 PoE injectors ship 100 Mbps ports — put a 100 Mb splitter behind a gigabit injector and you have built a 100 Mb link on both counts. Two more things worth knowing before you buy:

• Minimum load. Adafruit's page warns of “a minimum load requirement of at least 100mA on the 5V power supply”. A sensor board that idles under that can behave strangely on a splitter that a Pi runs fine.
• Vendor claims that contradict the board maker. Linovision's blog calls its 5 V 2.4 A, 12 W splitter “a perfect match for powering a Raspberry Pi 5” and “more than sufficient” for “any Raspberry Pi model, including the Raspberry Pi 5”. Raspberry Pi's documentation asks for 25 W on the same board. Both statements are published; they cannot both be a good buying guide.

What people actually run them on

The forum answer is consistently: a Pi with nothing plugged into it. In the Pi 5 + PoE HAT thread, ZSamuels28 described running Pi 5s off external splitters in February 2024 — “Both work fine with my Pi's, I assume if I start plugging in heavy external devices they'll be unable to power them” — and gave the reason in the next line: no peripherals, “they only pull 5-6W max”. On Hacker News, EvanAnderson calls splitters “cheap and very useful” while questioning a design that feeds power over USB-A, and hcfman reports PoE injecting noise into microphone inputs on a Pi, fixed only by adding a power-conditioning HAT on top. In a Home Assistant thread on small PoE splitters, the complaint is physical: the wall-box-sized version cost more than the tablet it was meant to power.

So which one

Match the splitter to the load, not to the biggest number on the box.

• A Pi 5 with an NVMe drive or powered USB devices. No USB-C splitter on this list will cover it honestly. Either power the board from a wall supply — the options are in my note on the Raspberry Pi 5 power supply alternatives — or go 802.3bt, where Linovision's 90 W POE-SP02BT-POE publishes 5 V at 14 A for $55 and hands it to you on a terminal block, not USB-C. The NVMe HAT census covers what that drive adds to the budget.
• A headless Pi 5 or Pi 4 with nothing attached. The 4 A class on an 802.3at source is the sane ceiling. The DSLRKIT 5 V 4 A gigabit unit sits there — its Amazon listing read In Stock on 2026-09-18. If you want the largest published rating anyway, and you read it as a peak rather than a promise, the Waveshare 5 V 5 A industrial unit is the one, also sold by PiShop at $29.95 as of 2026-09-18.
• A camera, an ESP32 panel, a small sensor board. An 802.3af 2 A unit like the Adafruit 4552 ($14.95, 100 Mb) is inside the standard's budget with room to spare — just check the minimum load and the port speed.
• Your switch is the variable. All of this assumes the source really is 802.3at. If yours is a passive injector or an af-only port, the splitter drops to its af rating, and only two of the twenty tell you what that is.

Boards that mount on the Pi instead of dangling off it are in the PoE HAT census. Across all three censuses the pattern repeats: the label describes the converter, not the link.

Sourcing for this post: the 20-splitter table was built by reading vendor specification pages directly — Waveshare's wiki, The Pi Hut, Adafruit, UCTRONICS, RAKwireless, 52Pi, POE Texas and Linovision — plus retail listings for the four models whose makers publish nothing else, all read on 2026-09-18. PROCET's site would not resolve for me, so its PD profile ladder is cited from its own product copy as it appears in search results and a distributor listing, and I have not linked it. The 12.95 W and 25.5 W figures come from TI application note SLVA221 and EDN respectively; the efficiency column is arithmetic on those two numbers, nothing more. Amazon prices are not quoted because my connection renders them in local currency; vendor prices are quoted with the date I read them. I own none of this hardware, so nothing here is a measurement of mine, and the operating experience is quoted from named posters with links to their posts.