On the AI/Gemini and the eventual replacement for an internal stack, Apple has done that before with Apple Maps.

At the start people laughed at the melting bridges and the airport in a farm (the popular Airfield farm in Dublin, which we visited countless times with our daughter and their friends), but, in the end, it's a competent replacement for Google Maps.

Apple is betting that good enough will get cheaper - with cheaper training, and that it will be possible to run good enough inference with local models fine tuned on the device with data you have on your iCloud. Google will still have their colossal structure and these huge deployments will, clearly, get us to superhuman levels of artificial intelligence, but that's a lot more than good enough.

As the MacBook Neo demonstrates, sometimes the brains of a phone is all you need for a desktop computer, and, if that's good enough for you, it makes no sense to get a Mac Studio with 256GB of memory, unless you want it to tune your iPhone's models in seconds rather than overnight on the charger.

My understanding was that Google was withholding access to features (vector tiles, turn by turn) to renegotiate maps usage. The two things I heard they wanted - features gated on Google login, and advertisements.

Apple had been developing their own maps solution already and accelerated it to land before the end of their existing agreement with Google, when Apple would have to accept Google's terms to continue usage. Google apparently had no idea negotiations had fallen through until the keynote.

It was going to ship as beta quality anyway (they were merging multiple sources of third party data), but the timeline meant it became a replacement in whatever form it was in.

Google themselves also did this with Gemini, we saw how quickly OpenAI's lead was erased. It could make quite a bit of sense to use this as a crutch so they (Apple) can overhaul their own efforts free of production pressures.

I'm not really seeing it though. Apple is more hardware and holistic experience, they aren't traditionally pushing the boundaries of "web-scale" nerd software. As long as they can keep on top of the devices that end up in people's hands they will have a place and don't need to be an AI powerhouse. They're doing just fine using Google Search.