After reading this, the reader knows what production increase was reported, why glass carriers matter and which numbers Samsung has not confirmed.

Samsung Reportedly Plans HBM4 Expansion

A Korean newspaper says Samsung will more than double advanced-memory output, but the company has not published the cited capacity targets.

Samsung Electronics is preparing a large expansion of high-bandwidth-memory production for next year, according to a report from Seoul Economic Daily that cites industry sources.

The newspaper says Samsung’s total HBM capacity is expected to reach about 250,000 wafers a month. It also reports that sixth- and seventh-generation products, HBM4 and HBM4E, would account for roughly 80% of that output. Outsourced glass-carrier volume would rise about 2.5-fold as part of the plan.

Those numbers have not been announced by Samsung. The article is original Korean reporting presented in an AI-translated English edition. It names no source for the forward production targets, so they should be treated as a reported plan rather than a confirmed commitment.

Why it matters

AI accelerators depend on HBM stacks placed close to the processor. The arrangement provides far more bandwidth than ordinary server memory, allowing large models to move weights and intermediate values quickly enough to keep expensive compute units busy. A shortage of qualified HBM can therefore limit accelerator shipments even when processor dies are available.

Samsung is one of only a few companies able to manufacture this memory at scale. More capacity from a second or third supplier can give accelerator designers additional sourcing options and reduce dependence on a single production line. It can also shift pricing and allocation across data-center customers.

HBM4 adds another manufacturing challenge. More memory dies must be thinned, stacked, connected and tested while meeting strict power and heat requirements. The reported emphasis on glass carriers reflects the packaging process: a stable carrier supports thin wafers and dies through high-temperature steps with less warping than some conventional materials.

The report says Samsung shipped mass-produced HBM4 from its Cheonan campus in February. The new claim is not that HBM4 exists, but that Samsung intends to increase the surrounding wafer and packaging capacity rapidly and move most output toward HBM4 and HBM4E.

Capacity is not qualified supply

A wafer target does not translate directly into usable accelerator modules. Yield, stacking success, packaging throughput and customer qualification all affect the final number of shippable products. An expansion can be delayed by equipment installation or by the time required to validate a changed process with a major customer.

The mix matters as well. “HBM capacity” can include different generations and stack configurations. A headline monthly wafer number does not reveal the number of memory stacks, their capacity, their speed or which customers have reserved them.

For buyers and investors, the most useful confirmation would be a Samsung capital-expenditure disclosure, supplier orders, factory capacity guidance or a named customer’s qualification announcement. Until then, the reported figures are a directional signal from one publication.

If the plan is carried out, it would widen the physical supply base behind AI systems. It would not automatically remove the bottleneck, lower prices or guarantee competitive yields. The story is meaningful because memory and packaging capacity can constrain the market; it is uncertain because the crucial targets remain sourced reporting.

Verification

  • Tier 1 — VERIFIED AS REPORTED: Seoul Economic Daily published the capacity, product-mix and glass-carrier targets on 20 September 2026. Source: https://en.sedaily.com/finance/2026/09/20/samsung-to-double-hbm4-output-next-year-sources-say
  • Tier 3 — UNVERIFIED: Samsung has not publicly confirmed the reported 250,000-wafer target, 80% HBM4/HBM4E mix or 2.5-fold carrier increase. Same secondary source.
  • Tier 1 — VERIFIED AS REPORTED: The article identifies Samsung’s February HBM4 mass-production shipment and Cheonan campus. Same source.
  • Tier 2 — ANALYSIS: Discussion of yields, qualification, supply and pricing is industry analysis rather than a forecast.

Glossary candidates

  • HBM: Stacked memory placed near a processor to provide very high data bandwidth.
  • Yield: The share of manufactured units that meet performance and reliability requirements.

Cold-reader sentence: Samsung may sharply expand HBM4 production next year, but its reported capacity targets still lack direct company confirmation.