NewsBite analysis · AI, Hardware, Startups
DigiTimes
Memory drives 62% of Nvidia Vera Rubin's cost — SOCAMM2, not HBM4, leads the bill
A recent analysis reveals that memory components account for approximately 62% of the costs associated with Nvidia's Vera Rubin AI platform. The report indicates that SOCAMM2, a type of memory chip, plays a more significant role than HBM4 in driving these costs. This insight highlights the current landscape of memory usage in AI technologies and suggests a shift in focus among chipmakers aiming to optimize costs. Moreover, industry experts predict that the memory shortage could transition into a surplus by 2028 as manufacturers ramp up production.
MonitorConfidence: HighMid term horizon
Published 11 Aug 2026, 00:20 UTC1 min read
Visit original article ↗Lead image — DigiTimes
Summary
TL;DR
- Nvidia's Vera Rubin AI platform costs are largely driven by memory.
- SOCAMM2 memory components accounted for the majority of costs, not HBM4.
- 62% of the platform's cost is attributed to memory components.
- Strong performance reported by Cambricon and Moore Threads in China's AI chip markets.
- Projected memory surplus by 2028 as production increases.
Interesting facts
- The substantial share of memory costs emphasizes the importance of supply chain management in AI development.
- Industry shifts may lead to new memory technologies vying for market dominance.