
Post
taishiro
The HBM trade is becoming a capacity allocation story, not just an AI demand story.
When memory inventory gets this tight, chipmakers face a choice: prioritize high margin HBM or supply conventional DRAM. That competition can push memory prices higher, but it also makes supply normalization the biggest risk to the trade.
SK Hynix has the advantage today, but Samsung's progress shows that leadership is not guaranteed.
#SamsungQ3ProfitOutlook
#SKhynixQ3HBM4Earnings
#Hynix Cloud +AI HBM4 Cloud = Memory bottleneck now, not GPU
HBM market $38B→$58B +52% 26E, $85B 27E. SK = 62% HBM share, *∼70% of NVDA Vera Rubin HBM4* (TrendForce) vs 50% est. 16 stacks = 576GB/stack vs HBM3E 8Gb → 2x BW.
Why bullish: Inventory <10 days (first ever). Wafers 150-200k/mo 3-4x Micron, 50% → HBM4 = DDR5 squeezed → Q4 DRAM +10-15% NAND +15-20%. JPM sees 20%
*Risks:* Samsung $8.7B HBM Q3 > SK $7.6B first time*
*OKX play: Long HBM leverage, short PC DDR.*

Ansvarsfriskrivning: OKX Orbit-innehåll tillhandahålls endast i informationssyfte. Läs mer
Svar
Inga kommentarer än. Var den första att svara!
Dagens marknadsrykte
1#Twenty One Capital ökade sitt innehav med 154 BTC, för första gången sedan juli 2025
Populärt
2#Nvidia siktar på rekordkvartalsintäkter på 100 miljarder dollar



3#Anthropic startar IPO-roadshow, värderingen stiger till 1,8 biljoner dollar

