10 GbE for local AI: link two boxes and a NAS for models
We earn from qualifying purchases via affiliate links. Disclosure.
Networking only matters for local AI when you store models on a NAS or link several AI boxes. A 4-port 10 GbE switch (check the current price at Amazon), a 10 GbE card for any desktop, and machines with built-in 10GbE or USB4 cover it. Splitting one model across machines with llama.cpp RPC is an experiment, and its own docs call it fragile and insecure.
Two jobs for a network
1. A NAS for model files. Model downloads run from a few GB to 81 GB (Qwen3.5 122B-A10B in Ollama). Keeping one copy on a NAS and loading it over 10 GbE saves downloading the same files on every machine.
2. Linking AI boxes. Minisforum's store page says the MS-S1 MAX can be clustered and has dual 10GbE and USB4 V2 at 80 Gbps. It also says two units in Minisforum's own testing ran a 235B-parameter model at Q4. That is the maker's claim; we have not checked it.
The llama.cpp RPC caveat
"This example and the RPC backend are currently in a proof-of-concept development stage. As such, the functionality is fragile and insecure. Never run the RPC server on an open network or in a sensitive environment!"
Picks
MikroTik CRS304-4XG-IN (4 × 10 GbE RJ45, fanless)
Price shown at Amazon
Sold at Amazon.
Check price at Amazon (opens in a new tab)
TP-Link TX401 10 GbE PCIe network card
Price shown at Amazon
Sold at Amazon.
Check price at Amazon (opens in a new tab)
Minisforum N5 Pro AI NAS, barebone (no RAM or drives; 128 GB OS storage)
$899 Minisforum · checked Oct 27, 2026, 12:40 PM CT
Caveat
Store data says estimated shipping mid-November for the HX PRO 375 version
Check price at Minisforum (opens in a new tab)
Minisforum MS-S1 MAX 128GB / 2TB (Ryzen AI Max+ 395)
$3,919 Minisforum · checked Oct 27, 2026, 12:40 PM CT
Caveat
Store data says estimated shipping late November. The 128GB / 4TB version was $4,199.
Check price at Minisforum (opens in a new tab)Not picked yet: 25 GbE switches and network cards. Price not checked.
Layout
- Decide what the network is for. Moving model files from a NAS, or splitting one model across machines. Copying files only needs fast storage and a fast link; splitting a model is an experiment.
- Give each machine a 10 GbE port. Mini PCs like the MS-S1 MAX have dual 10GbE built in, per the store page. A desktop can take a PCIe 10 GbE card.
- Connect them through a 10 GbE switch. A small 4-port 10 GbE switch covers two to four AI boxes plus a NAS.
- Keep the cluster on a private network. The llama.cpp RPC README says never run the RPC server on an open network or in a sensitive environment.
FAQ
Do I need 10 GbE for local AI?
Only if you move large model files between machines often, or experiment with splitting a model across machines. A single box needs no special networking.
Can I split one model across two PCs?
llama.cpp has an RPC backend for this. Its own README calls it a fragile, insecure proof of concept and says never to run it on an open network. We cover it as an experiment only.
What NAS works for storing models?
We list the Minisforum N5 Pro AI NAS (barebone, $899 on the Minisforum store on Oct 27, 2026). We have not used it.
