Even if they don't go that route, the data from icloud, cash on hand, and partnerships with sota labs, still position them as a frontier competitor that just hasn't launched yet.
Anyway you shake it strategically, Apple still owns the ecosystem end-to-end.
Apple mass-produces raster GPUs and bakes them into an expensive ARM SOC. They are not a drop-in replacement for CUDA or any of the expensive GPGPU hardware. Apple Silicon runs compute shaders just like the GPUs from 2012.
You could call them a "frontier competitor that just hasn't launched yet" but we don't even say that about AMD or Intel despite supporting Linux and shipping GPGPU architectures already. Apple's position in the greater AI/GPGPU industry is widely accepted to be forfeit. Even Google's AI hardware strategy is closer to the frontier than Apple's.
Apple Silicon is not used for inference in the majority of OpenClaw setups. macOS is popular for OpenClaw, but you can get the same results on a Linux, Windows or even ChromeOS client. There are OpenClaw desktops that run on a Raspberry Pi.
In the broader inference/training market, Apple Silicon is a non-entity compared to systems like the GB100.
It just happens to be popular among openclaw users who don't understand how it works. Which is the perfect target audience.
llms will be the new search and you'll delegate to them a lot, because it's so convenient (not saying it will be great!).