I’m still very bullish on Micron and memory in general , but only main concern I have around MU’s long term bull case is the thesis rests a lot on AI models needing more and more memory. When you look at what Kimi 3 and deepseek have managed to do with far less hardware power and performance, it makes me wonder if that assumption will continue to hold?
Chinese labs keep getting forced to optimise because they cannot just throw more accelerators at the problem, and the result is they are achieving great performance without the memory footprint that was assumed essential.. Then there’s CXMT getting their yields up and sampling HBM2 to domestic hyperscalers who would love to ditch the Western supply chain if they had the opportunity. I could be wrong though, but what are the chances of these chinese AI models slowing down memory growth?