Memory revenue is only up ~20-40%, but the memory makers' profits jumped an order of magnitude. What's actually driving the gap?

Trying to get my head around the memory cycle and something isn't adding up for me.

Total DRAM+NAND revenue went from \~$170B in 2024 to \~$200B in 2025, and maybe \~$290B in 2026. So call it up 20% then 40%. Solid, but not insane.

Meanwhile Micron's gross margin went from like 37% to 84% and it swung from losing money in 2023 to printing record profit. That's not a 40% improvement, that's a different universe.

I get the textbook answer, memory has huge fixed costs so price increases fall almost straight to the bottom line, plus they're coming off a 2023 trough where they were selling below cost. Operating leverage cuts both ways etc.

But here's what I can't settle: AI demand was already ramping in 2024. So why did the profit explosion wait until 2025-26? Was it just inventory finally clearing, or is HBM crowding out commodity DRAM capacity enough to drag the whole price deck up? And if the profit is basically all price, not volume, doesn't that make this way more fragile than the 84% margin makes it look?

Curious how people who've traded a few memory cycles think about separating the durable part from the cyclical part here.