SemiAnalysis: Memory's Share of Cloud Provider Capex to Exceed 40%, Market Not Fully Pricing In

BlockBeats News, July 4th, Semiconductor and AI research firm SemiAnalysis published an article pointing out that the proportion of memory in the capital expenditure of hyperscale cloud vendors is becoming a hot topic in the market. This especially gained attention after Micron's earnings report last week, with some market participants shocked by the high memory spending ratio expected next year. SemiAnalysis first proposed in February this year that memory spending would account for about 30% of cloud vendor capital expenditure. At that time, most clients questioned this number, as memory typically only accounts for low double-digit percentages of a server's BOM. In May, as memory prices rose faster than expected, SemiAnalysis further responded: by aggregating DRAM, NAND, and HBM, Nvidia's memory spending in systems will exceed 30% by the end of 2026 and further rise to over 40% in 2027.

SemiAnalysis predicts that the market's understanding of this structural change will gradually deepen in the coming months. This implies that amidst the continuous expansion of AI capital expenditure, the memory segment is transitioning from a "secondary" cost in server expenses to the "single largest cost item," with its impact on memory manufacturers' pricing power and profit margins not yet fully priced in.