I’m struggling with how to value semis when the earnings are real but probably not normal

I keep seeing the same argument around semiconductors: “it’s not a bubble because the earnings growth is real.”

I don’t totally disagree with that. NVDA, AMD, MU, and the rest of the AI hardware chain are not random 2021 SPACs with no revenue. There is real demand here. Data centers need compute, memory, networking, storage, power, all of it. This is not just a fake story.

But I’m not sure that answers the valuation question.

The part I keep coming back to is normalized earnings. With semis, current earnings can look very cheap right when the cycle is unusually favorable. Margins expand, pricing improves, operating leverage kicks in, and suddenly the P/E looks reasonable. Then people start treating that earnings base like it is the new floor.

Maybe this time is different in parts of the chain. NVDA clearly has a stronger moat than a typical cyclical chip company. But I’m less comfortable applying that same logic across every AI-adjacent name. Memory especially makes me nervous because shortage economics can look like pricing power until supply catches up.

That’s where MU is interesting to me. It can look cheap on forward numbers, but the real question is whether those numbers are mid-cycle, peak-cycle, or something genuinely structurally higher. I don’t have a clean answer. I just don’t want to fool myself into thinking a low multiple automatically means value if the “E” is doing most of the work.

This is also why I find the “AI bubble or not” debate kind of unhelpful. A stock can be tied to a real long-term trend and still be over-earning for a period of time. Both can be true.

For me, the question is not whether AI demand is real. It is how much of today’s margin and revenue strength survives if hyperscaler capex slows for a few quarters, or if supply finally starts catching up in the weaker parts of the chain.

I’m still trying to build a better way to think about this. Maybe the right answer is to value each part of the semiconductor stack very differently instead of throwing everything into the same AI bucket.

How are people here normalizing earnings for semis right now? Are you using current margins, a full-cycle average, or just avoiding the group because the cycle is too hard to handicap?