Samsung Electronics is expected to more than double output of its HBM4 and HBM4E memory in 2027, according to semiconductor industry sources cited by Seoul Economic Daily on September 20, 2026. The report’s central evidence is indirect: the company is said to plan an increase in outsourced cleaning of glass carriers to 50,000 sheets a month, from 20,000 a month in 2026. That projected 2.5-fold rise is being used as a proxy for growth in production of Samsung’s newest high-bandwidth-memory stacks.
The connection is plausible because these carriers perform a specific manufacturing function. They temporarily support an HBM wafer while it is ground thinner and drilled, helping prevent bending or cracking. HBM achieves high bandwidth by stacking several DRAM dies within a tightly constrained package height, so higher stack counts make controlled thinning and warpage management more demanding. Samsung’s planned HBM4-family expansion centers on products with 12 layers or more.
The reported carrier figures also show a sharp recent progression: requirements were about 10,000 sheets a month in 2025, doubled to 20,000 in 2026, and could reach 50,000 in 2027. Analysts cited by the report argue that this makes at least a twofold increase in HBM4-family output likely. That is an inference, however, not a disclosed production target. Glass carriers are reused after cleaning, and the number required per unit can vary with process loading, yields, and equipment utilization.
Samsung’s product timetable provides separate context. The company began mass-production shipments in February 2026 of HBM4 built with sixth-generation 1c DRAM and a 4-nanometer base die. In May, it supplied 12-layer HBM4E samples to customers that included Nvidia. These events establish that Samsung had moved HBM4 into shipments and was qualifying a denser HBM4E product; they do not independently confirm the forecast volume increase.
The displayed products provide visual context for Samsung’s sixth-generation HBM4, seventh-generation HBM4E, and plans for stacks with 12 layers or more. Credit: en.sedaily.com.
The shipment scene corresponds to the article’s discussion of Samsung beginning mass-production shipments of HBM4 made with 1c DRAM and a 4-nanometer base die. Credit: en.sedaily.com.
The wider production picture cited in the report is also based on industry expectations. Average monthly HBM wafer input is projected to rise nearly 40%, from roughly 180,000 wafers in 2026 to about 250,000 in 2027. Within that larger pool, the HBM4 family’s shipment mix is expected to increase from around 40% to about 80% as HBM4E enters mass production. Those figures suggest that Samsung’s expansion would involve both greater overall HBM capacity and a shift toward its newer, higher-value products, but neither projection is a Samsung commitment.
For buyers of servers and AI systems, the practical question is whether additional production will improve availability or merely follow already secured demand. The supplied evidence does not establish customer allocations, contract volumes, selling prices, or delivery schedules. It therefore cannot show that a larger Samsung output will quickly relieve supply constraints, lower accelerator costs, or make HBM4-based systems broadly accessible to consumers.
The carrier data is useful precisely because it exposes a less visible part of the HBM supply chain. Yet its diagnostic value has limits: increased cleaning volume can reflect higher starts, lower yields, changes in handling, or a different mix of products. The report supports a credible expectation of substantial HBM4-family growth, but only Samsung production data, customer qualification records, or verified supplier orders could turn that expectation into a measured outcome.