Samsung to double HBM4/HBM4E output as glass-carrier demand jumps 2.5x

Samsung is on track to more than double production of its HBM4-generation high-bandwidth memory next year, spanning both sixth-gen HBM4 and seventh-gen HBM4E. The clearest signal comes from an unusual proxy: glass carriers, the temporary glass supports bonded to wafers to keep them from bending or cracking while they’re ground thin and drilled. Samsung plans to lift outsourced cleaning of these carriers to 50,000 sheets a month from 20,000 today — itself up from just 10,000 a year ago. Even after accounting for the fact that carriers are reused and consumption varies by process, analysts read a 2.5x jump in carrier demand as strong evidence of at least a twofold rise in HBM4-family output.

The scaling centers on denser 12-layer-and-higher stacks, where warpage control and wafer-thinning become the binding constraints as die counts climb. Samsung began mass-production shipments of HBM4 in February, built on 1c (10nm-class) DRAM with a 4nm base die, and sent 12-layer HBM4E samples to customers including Nvidia in May. Overall HBM capacity is expected to grow nearly 40%, from roughly 180,000 to about 250,000 wafers of average monthly input.

The shift is also a mix change: the HBM4 family is projected to move from about 40% of shipments this year to roughly 80% next year as HBM4E ramps. That reflects a deliberate tilt toward higher-value parts, positioning Samsung to compete harder for the AI-accelerator memory market where HBM supply has become a key bottleneck.