I think it will be. Smarter, more efficient to inference, faster to inference. I don't think Apple will win any of these categories.

Therefore, I don't think they will be able to compete well in the new world.

Training frontier models is heinously expensive. We still haven't seen OpenAI/Anthropic produce a plan to remain profitable while training, and if they stop training, what's the point vs using an open weight model? They have to keep training in order to stay ahead.

You still haven’t given a single example of anything that an app would be replaced by that any normal user would want, ever. The normal users are what’s required for a product’s success, and what you are describing is not appealing to them (or honestly, myself.) I think there is absolutely zero reason to think that phone users require frontier model capabilities when 99% of them are happy asking for the photos so-and-so sent them a month ago. Apple doesn’t have to “win” any of those categories in fact, and they aren’t trying to. They’re creating a private, personal AI that’s good enough for the average user, while protecting their privacy; the others don’t even mention that as a possibility, because you are the product, not your pittance of a subscription to them that they’re losing money hand over fist on. Your argument completely forgets that economics is a thing.