SO you are saying we will need more constant memory, be indefinitely constraint, and memory will not be cyclical 🙂?
I just think it's more complicated than that. Yes, I think we will need more memory. Yes, I think it will be a continued shortage. But that doesn't mean price keeps going up. Like I said earlier...FY27 EPS makes it look stupid cheap. However, it's supposed to look cheap when earnings get inflated. When we get a cycle turn, earnings will be fall faster than price.
86% gross margin. That means every extra buck of price is profit. 50B quarter? 43B profit. Miss on blended ASP by 10% and that's 5B up in smoke. Several dollars of EPS in a single quarter. That can cause volatile gaps in price and the higher we ramp with estimates the more risk when the plane starts to rattle at high altitude.
I'm just saying...memory went from what...5 or 8% Capex spend to probably close to 40-50% now?
Hyperscaler "pacing", NVDA locking at lower memory counts in order to sell more GPUs, DRAM contract accelerations slowing or flattening out, there's several scenarios where the peak rate of price increases slows and starts getting behind you. With current margins that's what produces earnings drops, not saying that we now have a glut on memory.
Robots...sure, it seems very logical at first glance but these aren't data centers. 128GB DRAM + ~1TB NAND give or take? 50-80K robots over next year. Maybe 500K/yr by 2030? What's 500K robots x $800 memory = 400M. MU did ~42B in a single quarter.
I think robots....sure, will consume memory but that kind of demand is nowhere near what datacenters will demand. We're better off aiming for the land/orbital datacenter buildout which will consume way more memory.
I hope you're right on the MU earnings. Again, I'm still super bullish into next year but depending on what the stocks doing...I may trim at that point.