it got strong 'lady doth protest' energy
They will probably be the first to have mostly automated warehouses for general goods. So a lot of robotics and computer vision. Not as flashy as the new wave of generative models, but getting the robotics and models reliable enough for production will be valuable.
There is "AI" and then there is the AI we have today.
Alexa is in the same class as Watson when it comes to intelligence. You can reproduce most of its effects using old techniques on cheap hardware. NLP and statistical classification is incredibly powerful if you throw enough resources at it. These approaches are also mostly deterministic and easy to incrementally improve (to a limit - i.e. they can't generate arbitrary output).
LLMs are like enriching uranium by comparison. Unless someone at the top thought 'damn we need one of those things soon' a few years ago, there would not likely be much progress to talk about today.
I haven't ever associated Amazon with advanced ML research, or even a basic aptitude at using "AI" (heavy air quotes) to make their products better. I guess there is Alexa but that is largely a failure for anything other than basic tasks. It is def better than Siri but Siri must be a meme at this point because as I've said in a previous comment, it is beyond terrible.
I used to buy items from Amazon and I still get pages and pages of suggestions from these one off purchases that are so irrelevant, I think a 3rd year CS undergrad (if that) could write a better recommendations engine if given a user purchase history API and a product catalog API and legit nothing else