The biggest risk to memory stocks isn't collapsing AI demand; it's assuming today's exceptional profits will last forever. CXMT's 4F² architecture could improve memory density and cost efficiency, but commercial yields, mass production and server qualification remain unproven. Meanwhile, TrendForce still forecasts Q4 conventional DRAM contract prices rising 10–15%, suggesting near-term supply remains tight.

Here's the deeper issue: stocks move on future expectations, not just current earnings. When margins approach their peak, even record profits may fail to support valuations if investors anticipate tougher competition.

I'd watch DRAM pricing, production yields and HBM competitiveness rather than panic over one selloff. AI demand can keep growing while memory stocks fall. The real question is whether future earnings can justify today's prices.

@Tiger_comments [财迷]