[acdha]

On the other hand, servers are running more GPU / ML workloads these days and semiconductor designs are more automated. It’d be interesting to see if they could cost-effectively sell those designs with, perhaps, capacity tweaks similar to what they’ve done with the iPad X-series releases where the chip design is very similar but they’ve added extra GPU cores, cache, etc.

That’d also be consistent with their past history of trying to pick some edge for a particular market: optimized for media processing, ML, VR/AR, etc. rather than just competing with x86 for the generic workloads. I still don’t see it as a high probability but I’d expect a niche if it happens at all.