Right now everyone is still focused on NVDA when talking about AI, but I feel like if Agentic AI really takes off, the bottleneck might slowly shift from GPUs to CPUs and storage chips. This kind of AI isn’t just about chatting, it’s about planning, doing work, and calling tools, so CPU scheduling and memory read/write will become more and more important.

So I’ve been looking at AMD and MU lately. AMD not only has GPUs, but its EPYC server CPUs already have a solid foundation. If AI agents need more CPUs for coordination, this could definitely get revalued. MU is more direct – no matter who wins, storage chips are going to be essential. GPUs need HBM, CPUs need DDR, and MU is just collecting rent in the AI infrastructure.

Not saying to chase the high, wait for a pullback or earnings reports. What do you guys think, should we stick with NVDA for the second half of AI or is it time for AMD and MU to catch up?