Apple built custom servers and OS for its AI cloud Apple has revealed it created its own datacenter stack – servers using its in-house silicon and operating system – at its Worldwide Developer Conference (WWDC) on Monday. Cupertino hasn't actually announced the servers or OS (and never addressed rumors of its plan to make datacenter-grade processors). Instead, references to …
"The machines run a new operating system that Apple's described as "a hardened subset of the foundations of iOS and macOS tailored to support Large Language Model (LLM) inference workloads while presenting an extremely narrow attack surface."
So in other words its probably just tweaked FreeBSD or MacOS, as I doubt they would go to the trouble of writing something new from scratch when they already have both of these OS capable of running on their own ARM hardware.
In reality, only the CLI for MacOS is BSD. The underlying kernel which used to be referred to as Darwin is a mix of Mach with some non-microkernel components, which may owe some of their history to BSD, but probably not that much, and that which does is probably heavily modified from it's BSD roots.
Bearing in mind that the article says that much has been removed to reduce the attack surface, I would suggest that significant parts of the BSD tool set are not on this new OS.
What I wonder is what they've left in to allow identification of the inevitable system problems. Not being able to get metrics and access to a server is just asking for problems, <joke>especially if you have an AI going rogue!</joke>
《So in other words its probably just tweaked FreeBSD or MacOS,》
I was thinking a Darwin kernel stripped down to support just the restricted hardware of their "compute node" and their AI ASICs.
I would guess perhaps a basic hypervisor under that and/or secure containers under Darwin to run multiple instances although they seem to be treating the nodes like a hardware equivalent of a container.
Or perhaps its just a pile of PowerPC boxes. ;)
SEPOS (Secure Enclave Processor Operating System) runs on Apple's embedded Secure Enclave Processors and is based on the L4 microkernel. Similar to what they're claiming for their cloud systems, it's a very bare-bones environment designed to be resistant to all sorts of attacks (however, older versions have been broken due to an implementation error or two).
https://support.apple.com/guide/security/secure-enclave-sec59b0b31ff/web
I sat through the keynote and aside from the cringe-worthy antics at various points, plus the fact that several of the male presenters seemed to be sharing the exact same pair of shoes, it was all a bit so-so.
Apple AI, for the moment, seems to be an implementation of a chat LLM and Stable Diffusion so you can get help with correcting your writing and generating emoji on the fly. Hmm, not exactly earth shattering.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
