Data movement, not compute, is AI's real cost: near-memory investment map in 3 layers
FinanceYF5 · x · 2026-10-17
FinanceYF5 publishes the final episode of its AI data center series, focused on near-memory computing investment opportunities.
Core argument: the most expensive part of AI is moving data, not computing—reading memory once costs as much electricity as 6,400 chip additions.
The author splits the near-memory opportunity into three layers:
- Memory itself: most certain, but most exposed to cycles
- The middle layer: turning into a real business
- New approaches: still option-like bets
Framed as industry analysis, not investment advice.
More from Venture
- 98% of US Households Still Don't Pay for AI, Data Shows — FinanceYF5 · 2026-10-17
- Anthropic's Q3 revenue reportedly hits $16.8B, up 46% QoQ — Beth_Kindig · 2026-10-17
- Investor Julian Weisser launches Solo Angels, an angel network for solo-founded companies — julianweisser · 2026-10-17
- boxd SDK downloads jump 100x in two months, hitting 4,181 in a single day — JensHonack · 2026-10-17
- NYC Comptroller urges Anthropic to delay ~$2T IPO over AI safety concerns — GaryMarcus · 2026-10-17
- Broadcom: each gigawatt of AI compute could yield $30B in annual revenue for customers — Beth_Kindig · 2026-10-17