I can't speak to the impact on privacy, but the bulk of Google's AI features are implemented within the software, with the hardware helping accelerate those operations: With GrapheneOS installed, the 'AI' features that Google talks about will not be present by default.
The hardware that Google refers to when talking about the Pixel's machine learning / AI capabilities is the TPU (Tensor Processing Unit) hardware accelerators. Any app can communicate with it by way of the Neural Networks API to accelerate their operations, and it's properly isolated + respects app sandboxing. However, some Google apps access it directly in order to provide more acceleration. This doesn't provide any extra data to the apps, but it does add a bit of extra attack surface, so GrapheneOS has added a toggle to disable Google apps from directly accessing the TPU/GXP to reduce attack surface. With the direct access toggled off, operations that would normally make use of the TPU/GXP will be slower, i.e. Google Camera post-processing, Google Photos 'machine learning' related features, etc. This toggle would likely make its way to the Pixel 8 if/when supported, also.
All this talk about AI, machine learning, sending videos and other data to google servers for processing, overall made me nervous and I'm now not sure if I should take the plunge.
Where did you hear that Google would be sent your photos and videos for processing?
Would I be paying a lot of money for Pixel 8 features that I can't use with G.OS?
Any Google apps containing their respective 'AI' features can be installed on GrapheneOS if you so choose.