DRAM inventories have fallen to 2-4 week levels globally, the tightest supply conditions since post-COVID recovery, as chipmakers redirect production toward high-bandwidth memory (HBM) for AI servers. The shift is creating constraints across international semiconductor markets as hyperscalers in North America, Asia, and Europe accelerate data center buildouts.
Memory manufacturers worldwide are reallocating production capacity from traditional DRAM to HBM, which commands premium pricing but requires specialized manufacturing processes. The reallocation is pulling resources from standard memory lines to meet demand from GPU clusters and AI accelerators deployed by cloud providers across continents.
US-based Analog Devices reported strong demand from global data center customers as AI infrastructure spending continues. The company cited industrial and data center segments as key growth drivers, reflecting sustained strength across international markets.
Traditional DRAM markets face capacity concerns as manufacturers balance commodity demand against AI-specific requirements. Industry observers note diverging patterns between standard memory chips and AI-optimized variants, with implications for pricing across global supply chains.
Cisco introduced its Silicon One G300 networking chip to support AI-scale deployments worldwide. "AI at scale demands open, standards-based networking that customers can deploy with confidence across diverse environments," said Yousuf Khan, highlighting infrastructure needs beyond memory.
Cirrus Logic forecasted Q4 FY26 GAAP gross margin between 51-53%, reflecting broader semiconductor pricing dynamics. SiTime's acquisition of Renesas' timing business—spanning operations in Japan and beyond—is expected to boost earnings in year one, showing consolidation as companies position for global AI infrastructure growth.
Texas Instruments' acquisition of Silicon Labs, which achieved 15% compound annual revenue growth since 2014 across international markets, demonstrates industry reshaping around AI demand.
The memory cycle split creates challenges for chipmakers managing production capacity across global fabs. HBM requires specialized processes concentrated in advanced Asian foundries, while traditional DRAM faces margin pressure from oversupply risks as capacity additions come online in Korea, Taiwan, and China.
Analog and specialty chip makers report sustained strength from AI data center buildouts worldwide, contrasting with cyclical concerns in commodity memory markets. The divergence signals a structural shift in global semiconductor demand as AI infrastructure becomes the dominant growth driver.