The author only mentions homomorphic encryption in a footnote:

Notes:

(A quick note: some will suggest that Apple should use fully-homomorphic encryption [FHE] for this calculation, so the private data can remain encrypted. This is theoretically possible, but unlikely to be practical. The best FHE schemes we have today really only work for evaluating very tiny ML models, of the sort that would be practical to run on a weak client device. While schemes will get better and hardware will too, I suspect this barrier will exist for a long time to come.)

And yet Apple claims to be using homomorphic encryption to provide their "private server" AI compute:

Combining Machine Learning and Homomorphic Encryption in the Apple Ecosystem

Presumably the author doubts Apple's implementation but for some reason has written a whole blog post about AI and encryption and hasn't mentioned why Apple's homomorphic encryption system doesn't work.

I'd be quite interested to know what exactly is the weakness in their implementation. I imagine Apple and everyone who uses their services would be interested to know too. So why not mention it at all?