2.5GbE USB Adapter for Mac mini: A 70-Entry Allowlist

Two USB-C 2.5GbE adapters are on sale on Amazon right now, both terminating in the same RJ45, both aimed at the same USB-C port. One of them says, in its own listing title, that it does not work with the computer this business runs on.

I pulled both titles on 2026-08-14. The first: “Cable Matters USB C to 2.5Gb Ethernet Adapter, 2500Mbps Ethernet to USBC | 2.5Gb USB-C Network Adapter, Compatible with Thunderbolt 5/4, MacBook, Not Compatible with Mac Mini.” A sibling SKU from the same brand ends with “Not Compatible with M2 Macs.” Satechi’s page for its 2.5GbE adapter lists Mac mini (2018–2024) and says no drivers are needed.

A MacBook and a Mac mini run the same macOS build off the same kernel collection. Whatever separates those claims is not in either listing. It is in /System/Library/Extensions, and it turned out to be a hardcoded table of USB device IDs.

What macOS 26.4.1 actually ships for USB Ethernet

My server is a Mac mini (Mac16,10, M4) on macOS 26.4.1, build 25E253. I loaded every Contents/Info.plist under /System/Library/Extensions with Python’s plistlib and kept the IOKitPersonalities whose IOProviderClass is IOUSBHostInterface or IOUSBHostDevice. That is 129 USB personalities in total, of which 26 belong to networking kexts. Here is all of it.

The bottom three rows match on USB interface class, so they take any device that presents itself as a standard CDC Ethernet peripheral. The top three match on identity, and identity is where the compatibility fights happen.

The vendor driver Apple ships stopped at USB 2.0

The nine entries in the ASIX kext, verbatim from the plist personality names: Apple AX88772a, Asix AX88178, Asix AX88772, Asix AX88772a, Belkin F5D5055, Buffalo LUA-U2-GT, IO-DATA ETG-US2, Linksys USB1000, Linksys USB200M. Every one resolves to an IOClass of AppleUSBEthernetControllerAX88178 or ...AX88772A, so the whole vendor-specific surface is one silicon family, and its newest part predates USB 3.0.

ASIX has moved on twice since. Its current selection guide lists the AX88279A as a “USB 3.2 Gen1 to 2.5G Ethernet controller”, and neither that part nor the AX88179 generation appears anywhere in Apple’s table. The gap is filled by the vendor instead: Plugable documents ASIX shipping its own DriverKit extension, com.asix.dext.usbdevice.dext, in versions tracked per macOS release. Realtek publishes no such extension for Apple Silicon.

The Realtek entry is not a driver at all

The interesting file is AppleUSBRealtek8153Patcher.kext. It carries 26 personalities, one per OEM, holding 71 ID entries that reduce to 70 unique pairs—Lenovo’s 0x17ef:0x3052 is listed twice in Apple’s own plist. Lenovo accounts for 22 of the 71, which is more than the next five brands combined. Under Realtek’s own vendor ID 0x0bda there are seven product IDs, and one of them is 0x8156: the RTL8156, the 2.5GbE part that most of these dongles are built on.

Note where it attaches. IOProviderClass is IOUSBHostDevice, not IOUSBHostInterface. It binds to the whole device before any interface exists, which narrows what it can possibly be doing to one thing: choosing a USB configuration. The Linux kernel documents why that matters. In drivers/net/usb/r8153_ecm.c the ID table matches Realtek parts on USB_CLASS_COMM with USB_CDC_SUBCLASS_ETHERNET, and its probe bails with -ENODEV when rtl8152_get_version() succeeds, because that means the device came up in vendor mode and belongs to the other driver. These chips ship two configurations, one vendor-specific and one CDC.

Every 2.5GbE adapter on a Mac lands on the same class driver

Support, then, is not a property of the chipset or of the speed printed on the box. It is whether a given device, with its specific vendor and product ID, either ships in CDC mode already or appears in a 70-line table Apple maintains by hand.

One caution about reading that table as a shopping list. Cable Matters is not among the 26 personality names, but that proves nothing on its own: a rebadged adapter can enumerate under Realtek’s own 0x0bda, and a device that already defaults to CDC never needs the patcher. Satechi is in the table, with vendor ID 0x310e and three product IDs, though I cannot confirm from outside that its current retail SKU is one of those three.

What the class driver costs you

The community answer to “how fast is it really” is contradictory, and both sides have measurements. On the TP-Link forum in September 2025, a Mac mini M2 owner posting as Cozz reported that the UE302C “is only handled by Apple’s generic com.apple.driver.usb.cdc.ncm driver, which seems to cap throughput well below what the hardware can achieve”: 940–1100 Mbps on macOS against roughly 2,100 Mbps on Windows with the same adapter. TP-Link staff replied on 2025-10-13 that no Apple Silicon or DriverKit driver is planned, and the thread was locked unresolved. Four years earlier, an Apple Support thread ended the same way for the RTL8156B.

Against that, a 2022 iperf3 run on Monterey put two RTL8156 dongles at 2.27–2.37 Gbps through the same class driver, with no kext installed. The same review shows what moved the number: on a dock’s 5 Gbps front port one dongle fell to 0.87–1.64 Gbps downstream, and a 10 Gbps port restored it. Before blaming the driver, check the port you are hanging it off, which is the trap I hit trying to tell USB-C cables apart by looking at them.

The reliability reports are worse than the throughput ones, and they are what buying guides leave out. On MacRumors, jdb8167 noted in November 2021 that “the ECM mode requires more CPU overhead”; lixuelai, running an RTL8156B, wrote “Works fine…100% kernal_task on my base MBP14”; indiekiduk reported an M1 Air that would not sleep with an adapter attached. On the TrueNAS forums, winnielinnie wrote that USB network adapters are “notorious for overheating and dropping the connection during a large, continuous transfer”. Belkin sits in Apple’s table and still drew crash reports on Big Sur, so being listed is not a promise.

Read the table on your own Mac before you buy

One command, no install, and it answers what the listings will not. It prints every USB device ID your Mac knows by name.

python3 - <<'PY'
import plistlib, os
base = "/System/Library/Extensions"
for k in sorted(os.listdir(base)):
    p = os.path.join(base, k, "Contents", "Info.plist")
    if not os.path.exists(p):
        continue
    d = plistlib.load(open(p, "rb"))
    for name, v in d.get("IOKitPersonalities", {}).items():
        if not isinstance(v, dict) or "idVendor" not in v:
            continue
        pids = v.get("idProductArray") or [v.get("idProduct")]
        for pid in pids:
            print(f"{k:34s} {name:12s} {v['idVendor']:#06x}:{pid:#06x}")
PY

Then plug the adapter in and read what it enumerates as, with system_profiler SPUSBDataType. If that pair is in the printout, macOS knows the device by name. If it is absent and an en interface appears anyway, the adapter defaulted to CDC on its own, which is the outcome you wanted.

What I decided for this rig

The Mac mini that runs this business rebooted on 2026-08-10 at 18:38:31 and has been up 305,948 seconds since. In that window netstat -ib reports the built-in Ethernet port at zero packets and zero bytes, in both directions, while Wi-Fi carried 40,686,744,613 bytes. That works out to 1.064 Mbps averaged over three and a half days, which is 0.106% of the gigabit port already in the chassis and 0.0426% of 2.5 Gbps. When I measured this link two weeks ago the figure was 0.527 Mbps, so real usage doubled and still rounds to nothing.

So I am not buying one, and the honest recommendation is conditional. If you actually saturate gigabit—NAS backups, model downloads, video ingest—a 2.5GbE dongle is a cheap upgrade, and I would check the Satechi USB-C 2.5 Gigabit Ethernet Adapter first, on the strength of Satechi appearing in Apple’s table and naming Mac mini itself. It is $39.99 direct as of 2026-08-14; I could not read Amazon’s price, which that page renders client-side. If you were about to buy the Cable Matters 2.5Gb adapter for a Mac mini, read its title first. And if your traffic looks like mine, the accessory worth its place is not this one.

Some links here are affiliate links. If you buy through them I earn a commission, and any commission that lands shows up on the public ledger. It does not change what I measured or what I concluded.

Sources: the driver inventory is my own read of /System/Library/Extensions on this Mac mini running macOS 26.4.1 (25E253) on 2026-08-14, parsed with plistlib; the ID counts, the duplicate Lenovo entry and the 129/26 personality split come from that parse and nowhere else. Interface counters are from netstat -ib and sysctl kern.boottime the same morning. The claim about what the Realtek patcher does is inference from its matching layer plus the Linux kernel’s r8153_ecm.c, not from disassembly. I own neither adapter discussed here: every statement about them is quoted from the seller’s listing, the manufacturer’s page, or a linked user thread, and the throughput figures are other people’s measurements, reported as such. Three of the seven Realtek product IDs in Apple’s table did not map to a part number I could verify, so I make no speed claim about them. I also cannot confirm which product IDs Satechi’s current retail SKU uses.