Samsung to Double HBM4 Memory Chip Production for AI Boom
Glass carrier demand surges 2.5x as Samsung prepares 12-layer HBM4E stacks for Nvidia.
1HBM4 output doubling
Samsung plans to more than double production of its HBM4 and HBM4E DRAM chips in 2025, responding to surging demand for high-bandwidth memory in AI accelerators. The company will increase monthly glass carrier usage from 20,000 sheets this year to 50,000 next year. These carriers prevent wafer damage during thinning processes essential for multi-layer HBM stacks. The 2.5x carrier increase strongly indicates at least 2x HBM4 output growth.
212-layer stack focus
Samsung's production ramp centers on 12-layer and higher HBM4E stacks, requiring advanced wafer thinning technology. As stack counts increase, controlling warpage becomes critical - each die must be precisely aligned within strict thickness limits. The company began mass-producing 10nm-class HBM4 in February and delivered 12-layer HBM4E samples to Nvidia in May, positioning itself for the next AI chip generation.
3Glass carrier criticality
Glass carriers serve as temporary supports during HBM wafer thinning, preventing bending or cracking when DRAM is ground to ultra-thin profiles. While carriers are reused after cleaning, their consumption still directly correlates with HBM output volume. From 10,000 monthly sheets in 2023 to a projected 50,000 in 2025, this 5x increase over two years underscores the production scaling challenge.
4Market share strategy
Samsung appears to be prioritizing high-value HBM4 products over older generations, with the HBM4 family projected to grow from 40% to 80% of shipments. The company aims to produce approximately 250,000 HBM wafers monthly in 2025, up 40% from 180,000 this year. This strategic focus comes as competitors like SK Hynix also accelerate AI memory production.
5Process technology edge
Samsung's HBM4 combines 10nm-class sixth-generation DRAM with 4nm-process base dies, while HBM4E pushes stacking technology further. The thinning and alignment precision required for 12+ layers represents a significant manufacturing hurdle. Industry analysts note yield variations across different process loading methods add complexity to output projections.
6AI memory arms race
The HBM production surge reflects intensifying competition to supply AI accelerator chips, where memory bandwidth often bottlenecks performance. Samsung's moves counter SK Hynix's early lead in HBM3 for Nvidia GPUs. With HBM4E sampling already underway, the company is positioning for next-gen AI hardware requiring even higher bandwidth memory configurations.
7Supply chain impact
Samsung's output expansion will significantly increase outsourced cleaning volumes for glass carriers, affecting specialty materials suppliers. The 2.5x carrier demand growth exceeds the expected HBM output increase due to higher-layer stacks requiring more processing steps. This ripple effect demonstrates how advanced memory production impacts multiple supply chain tiers.
8Memory market shift
Samsung's HBM4 focus marks a strategic pivot toward high-margin AI memory as traditional DRAM margins compress. The 80% HBM4 mix target for 2025 suggests rapid phase-out of older generations. This transition mirrors broader semiconductor industry trends where AI-driven architectures are reshaping production priorities and capital allocations.