Apples plan has been pretty obvious. They invested in small locally running features that provide small utility rather than massive hosted models that cost a fortune and aren’t profitable.
There also doesn’t seem to be much risk in falling behind. If you wait longer you can skip buying the now obsolete GPUs and training the now obsolete models.
They invested a ton in their Private Cloud Compute though, but are barely using their capacity.
https://9to5mac.com/2026/03/02/some-apple-ai-servers-are-rep...