I think the chances they’re going to make it an iPhone 17 feature and not support it on the iPhone 16 are very low. There’s nothing indicating the issues causing the delays they’re running into have anything to do with RAM. 8GB is plenty of RAM, especially with how efficient iOS is on RAM usage. Many full fledged computers use that RAM spec, including even Macs until recently. They’ve likely just run into major bugs affecting functionality, or security vulnerabilities that need fixed before release. That seems to be the consensus of everyone more familiar with the matter…Their roadblock will be getting Siri 2.0 to work within the parameters of their current on-device LLM. If they were to offload it to a server, even a PCC one they'd have no excuse not rolling it out to older devices which would remove their advertised upgrade path. This will never happen.
My guess is they are waiting for future iPhone models with more RAM to run them and it will not be an iOS19 feature but rather an iPhone 17 one (or whatever). But they've already advertised it to iPhone 16 buyers! Releasing it exclusively for the iPhone 17 or later will no doubt result in a few 'not as advertised' lawsuits.
Given the market apathy towards LLMs in general, I do wonder if they shouldn't just tank the whole idea and instead look at other areas of the OS or apps that might benefit from the current AI backend instead. Perhaps a presentation generator for Keynote or an 'Audio Playground' app to create alert tones or garageband loops from a hum or recording (and kill off the little-used Tone Store in the process). I imagine a sweet piece of marketing where a parent creates a Homekit alert using the cry from their newborn as the source or something 💕