Ya, these high memory amounts and ever increasing memory bandwidth are heavily (but not only) targeting people wanting to run local large AI models like a full deepseek on their machines.

You might not be able to train as well on them as NVIDIA + CUDA, but for local inference, they're an alternative to NVIDIA and more reasonably priced for the model sizes you can run, and each iteration they get better as the bandwidth increases.

I slowly turn to dust as I recall cracking open 2013 MacBook Pros and just putting more memory in.

The memories of loading up the G3 with SDRAM so I can fiddle with Photoshop 5, lost like tears in the rain.

https://feddit.org/post/32576427

Their DDR5 chips are 30% more expensive, but they are willing to sell 10% below market rates.

The article says that they won't catch up in 2027 but I expect China to pour in resources to catch that opportunity.

But I bet next they will close it after gaining bigger foothold. I wonder how would you prevent this.