10GbE Adapter for Mac: macOS Matches 14 of 25 Aquantia IDs

The Mac mini that runs this business has one Ethernet port, and system_profiler SPEthernetDataType says it is a Broadcom part: vendor 0x14e4, device 0x1682, driven by AppleBCM5701Ethernet. That port tops out at 1 Gbps. Apple's 10GbE build-to-order option would put a different chip on the board, and every Thunderbolt 10GbE adapter on the shelf carries a chip of its own. Whether that chip works the moment you plug it in depends on whether its PCI ID appears in a table macOS ships with.

So "best 10GbE adapter for Mac" is a question about chips, not brands. My server is a Mac mini (Mac16,10, M4) on macOS 26.4.1, build 25E253. I loaded every Info.plist under /System/Library/Extensions and /System/Library/DriverExtensions and kept the personalities that attach to a PCI device and belong to a network driver. Then I matched eleven adapters and setups to their controllers from vendor spec pages, read 2026-09-19. I own none of the adapters; what I say about them is a vendor statement or an owner report.

What macOS 26.4.1 ships for PCIe Ethernet

A Thunderbolt Ethernet adapter is a PCIe network card behind a Thunderbolt bridge, so macOS sees the card, not the adapter. Every personality below has IOPCITunnelCompatible set to true, which lets it load behind a Thunderbolt tunnel.

The four DriverKit extensions match on vendor alone and decide inside the binary, so their device lists aren't in the plist. I read them from the strings in each binary instead. The E1000 extension carries 179 "Intel(R) ..." model names, and none of them is an I226. The IXGBE binary has only one generic model string, so what I can report there is function-name families: 39 functions for the 82599, 19 for the X540, 20 for the X550. It also carries Intel's "Unsupported SFP+ module type was detected" error, which matters if you plan to use third-party optics in an Intel SFP+ adapter. That reading of the binary shows the code paths exist. It does not prove every card in those families initializes.

Aquantia: 14 of the 25 IDs Linux knows

Most compact 10GbE adapters for a Mac use a Marvell (formerly Aquantia) chip, and so does at least one of Apple's own 10GbE ports: ServeTheHome found an "Apple AQC113" in the M1 Mac mini's 10GbE option and measured 9.4 to 9.5 Gbps. I have not confirmed which chip the M4 option uses. For a reference list I used the Linux atlantic driver's aq_common.h, which defines 25 device IDs. The macOS kext matches 14 of them.

The pattern is regular. For the AQC10x generation, macOS lists all six "S" variants (AQC100S through AQC112S) but only one plain chip, the AQC107. The plain AQC100, AQC108, AQC109, AQC111 and AQC112 aren't matched. For the AQC113 generation, it lists the AQC113, AQC113CS, AQC113CA and AQC114CS, and leaves out the AQC113C, AQC115C and AQC116C.

Being in the table is necessary, not sufficient. In 2017, when this driver first shipped, the developer Mieze looked at why an ASUS XG-C100C (plain AQC107) failed and found the kext checks the card's subsystem vendor and ID. The workaround was a binary patch to the kext, which is not something you can do on an Apple silicon Mac without lowering its security settings. An owner on ServeTheHome put two AQC107 PCIe cards, including an ASUS Areion 10G, in a Sonnet Thunderbolt enclosure in 2022, and "neither worked at all" on an M1 Max. In the same thread another owner reported Sonnet Solo10G SFP+ units working on M1 Max and Ultra Mac Studios. So a generic PCIe card in an enclosure is a gamble even when its chip name looks right. The adapters built for Macs are the safer bet.

Eleven adapters and setups, by chip

Chips come from Sonnet's spec pages (for example the Solo10G), QNAP's hardware specs, OWC's support manual and the WAVLINK listing. AKiTiO's driver page says its driver is "compatible only with Intel-based Macs," which rules that one out for Apple silicon.

The two budget USB4 adapters are the weak spot in the table. Their listings say "AQC113" without the suffix. If a unit ships with an AQC113C (0x14C0), that ID isn't in the macOS table. Owners on Hacker News do report the IOCREST version working on Macs, but I found no one who posted the PCI ID it reports. If you buy one, check system_profiler SPEthernetDataType on the day it arrives, inside the return window.

What owners complain about

Heat is the complaint that comes up most. One owner who uses the IOCREST USB4 AQC113 adapter on Macs wrote that it "gets uncomfortably hot" while sustaining 9.3 Gbps. Another reported an IOCREST AQC113 card running at "like 90°C even with the heatsink." Owners who moved to an SFP+ Thunderbolt dongle on fiber say it runs considerably cooler than copper 10GBASE-T, which fits what I found on SFP+ to RJ45 module power.

The cheap Realtek alternatives fall short on Macs. The RTL8159 USB adapters are USB 3.2 2×2 devices, and Macs don't support 2×2, so they fall back to 10 Gbps signalling. The same owner measured about 6 Gbps with jitter from a MacBook Pro, and later put the ceiling at 7 Gbps. The RTL8127 PCIe chip has no Apple driver that I could confirm. The AppleEthernetRL mask spans 0x8124–0x8127, but its only personality is named RTL8125, and the one project that has an RTL8127 running on Apple silicon, RTL812xLucy, ships its own DriverKit extension. Its SFP+ variant, 10ec:0e10, is outside the mask entirely.

There are smaller limits too. Sonnet's spec pages cap jumbo frames at 9000 bytes under macOS, against 16352 on other systems for the Solo10G SFP+. For the MLX5 route, KittenLabs ran a ConnectX-6 in a TBU401 enclosure and got 20,467 Mbit/s down but 5,670 up, and the card needed an external 12 V supply because the port's 15 W was not enough. On Linux, one owner had both AQC100S adapters, the QNAP and the Sonnet SFP+, fail under Ubuntu LTS. Not a Mac problem, but it matters if the adapter moves between machines.

Which one, on paper

For a Mac on a 10GBASE-T switch, the safest picks are the ones whose chip suffix you can read on the spec sheet and match to the table: Sonnet Solo10G (AQC107S) or OWC Thunderbolt 4 10G (AQC113). When I checked Amazon for a New York address on 2026-09-19, none of the Solo10G, OWC or QNAP T310G1T listings had a featured offer, so I'm not linking them. Buy from the maker. The Amazon listing titled "Sonnet Solo10G SFP+" had bullet points describing the RJ45 model, so read the box before you click buy there.

As an Amazon Associate I earn from qualifying purchases. The cheapest in-stock option was the WAVLINK USB4 10G adapter at $139.99 with a 3.9-star rating from 38 ratings, and it comes with the chip-suffix caveat above. If you need two ports or SFP+ with an Intel chip, the Sonnet Twin10G was $569.99 with nine left, and it needs its own 60 W power supply. Commissions land on the public ledger.

If 2.5 Gbps is enough, the USB side has its own 2.5GbE adapter allowlist, which I read the same way, or a Mac mini dock with 2.5GbE. At 10 Gbps you also need the switch and cable: 2.5GbE switches with a 10G uplink and Cat6 vs Cat6A.

Update 2026-09-19: I ran the same plist read over macOS's serial drivers. The USB to serial adapter for Mac post has the result: 106 matched IDs against Linux's 1,162, and Prolific's whole PL2303 G series is missing.

Method: the driver tables come from my Mac mini (Mac16,10) on macOS 26.4.1 (25E253), read 2026-09-19 with Python's plistlib over every Info.plist in /System/Library/Extensions and /System/Library/DriverExtensions whose personality attaches to IOPCIDevice; DriverKit device coverage is inferred from strings in the binaries, which shows code paths, not tested hardware. The Aquantia reference list is Linux master's aq_common.h, fetched the same day. Chips for the eleven entries come from vendor spec pages, one Amazon listing and AKiTiO's driver page; owner reports come from ServeTheHome's forum, InsanelyMac, KittenLabs and six Hacker News comments read through the Algolia API. Amazon prices and stock were read with a New York delivery address on 2026-09-19. I own none of the adapters, and the only hardware I measured is this Mac mini's built-in port.