"Almost certainly" a bubble. That's Citadel's Ken Griffin on AI this week, bec… · @Leoskie_L · TwKit

@Leoskie_L

"Almost certainly" a bubble. That's Citadel's Ken Griffin on AI this week, because companies struggle to turn AI spend into productivity (via @KobeissiLetter). He also said he's long-term bullish. I'll skip the productivity fight. Bubbles usually burst after a glut shows up. In 2000 the glaring one was fiber. On the physical side today, it's the reverse. Compute: Musk on the SpaceX Q3 call, asking around to rent inference capacity: "What about some GB300s, don't have any of those. What about some GB 200s, don't have any of those. What about some TPUs? … how about some H100s." (Motley Fool transcript, unofficial) Memory: TrendForce (10/20), citing Commercial Times, reports institutional investors expect Q4 DRAM contract prices up ~20% QoQ, after ~80% in Q2 and ~40% in Q3. Slowing gains, not falling prices. Substrates: Unimicron's chairman says the most critical materials could be ~20% short in 2027, depending on actual demand, with 80%+ of tools on longer lead times (Economic Daily News, 10/20). Demand: Prismark sees the 2026 IC substrate market at ~$21.5B, +44% YoY (via @semivision_tw). That's growth, not a shortfall figure. It's like calling a housing bubble while builders queue for rebar and cement. Prices may be too high; it's hard to say too many houses were built. For the bubble call to stick, I want to see some layer with idle tools or piling inventory. I haven't found it yet. Maybe I just haven't looked in the right place. Not investment advice.

曝光 154 · 评论 0 · 点赞 0 · 书签 0