Yeah, the reason Ampere lost so much value is everyone else is doing what they are, especially Nvidia, whom also has the AI benefits.
I actually think AMD might make more sense for that. Maybe (AMD did buy that company years back that was I think using ARM designs, but think they'd bought them for their "fabric" to help with their Zen enterprise stuff, not their ARM expertise). Mostly because AMD needs to be offering ARM designs. And AMD seems to be in acquisition mode as they pivot their entire business towards datacenter. But I'm not sure AMD would even need them though.
I feel like the future is gonna be hybrid designs, x86 as control chips, paired with high core count efficient dense ARM for the core count parallel tasks, but then its probably already too late for AMD to make such a change though as ARM looks like its already winning people over for client/consumer designs due to the usability benefits (battery life), and the datacenters can just put x86 servers as directors (think they're already doing this?). Maybe there's still time but AMD would need the designs already started. Ditch the ZenXc stuff, and pair like 8 core x86 CPU chiplet with like a 16+ (not sure on the density they could go for the ARM chips, could they do 24, 32 for relative size) ARM chiplet for parallel tasks. Being fair, I think Microsoft should take the blame there, as if they'd been pushing for hybrid like that, it would've helped ARM on Windows adoption, without sabotaging other performance (like gaming for instance), and it would've made datacenters more efficient (plus the density would've pushed more core counts quicker). And like how in GPU, compute tasks that just need lots of parallel cores seem less dependent on the x86/ARM difference so software compatibility would've been less of an issue, which was the biggest problem with Microsoft's attempt at making Windows on ARM devices.
I don't think nvidia would want this, they're already making monster-sized arm chips in the datacenter. Article mentions Oracle as one of Ampere's key customers, maybe they'll just buy Ampere outright.
I actually think the overall reason for Ampere doing this is why that doesn't make sense either. That entire market/industry has pivoted so strongly to AI that they have no real value when you can just buy integrated Nvidia AI boxes offering what Ampere does, which they are gonna be doing anyways for the AI processing. Oracle is advertising their cloud service explicitly to boost AI for those that want it but don't have the resources to actually build out that.
I guess more traditional ARM based server chips might help them keep offering those services (or include them in new datacenters) but at lower footprint/power (so most of their buildout can be focused on AI), but can't imagine that has much value (as again, they could just do that on the Nvidia stuff when/if its not being utilized for AI). Or maybe going Ampere combined with Nvidia dGPU/AI add in cards would let them do more efficient AI buildouts that could also be used for other.