Memory is 70-80% of AI rack silicon, and Rasgon asks what if prices never collapse

firstadopter · x · 2026-10-25

Key takeaways from a CASPA panel with Bernstein's Stacy Rasgon and SemiAnalysis' Dylan Patel:

Memory is the big variable: Rasgon notes memory makes up 70-80% of an AI rack's silicon and AI memory consumes 3-4x the wafers per GB vs standard DRAM. Memory stocks trade at just 2-4x forward earnings on collapse fears — he asks "what happens if pricing never collapses?"
Substrates are the top bottleneck: Patel ranks substrates as "the biggest bottleneck of all," and argues power matters less than assumed — nearly 50GW of data centers are being built globally next year.
Financing signals: hyperscalers, "the most cash-generative businesses in the history of humankind," are now burning cash; neocloud debt backed by lab offtake prices at 13-15%.
Tool vendors face shortages of simple parts.

More from Venture