Samsung Says HBM Could Take Nearly 30% of DRAM Wafer Capacity in 2027

High-bandwidth memory could account for nearly 30% of DRAM makers' total wafer capacity in 2027, up from roughly 20% today, according to Samsung Electronics Executive Vice President Kim Taewoo. The September 29 estimate describes an industry-wide manufacturing shift rather than only Samsung's own output.

Kim said HBM and conventional DRAM draw on the same front-end wafer capacity, so expanding HBM production can restrict capacity available for standard DRAM. That connects the AI-memory buildout to the broader memory market even though HBM itself is not ordinary PC system memory.

Wafer-capacity share is a manufacturing allocation metric. It does not mean HBM will represent 30% of DRAM bit shipments, unit shipments or industry revenue. HBM stacks multiple DRAM dies and also requires advanced packaging, test and base-die work.

The estimate is still useful because wafer starts are finite. Directing more front-end capacity toward HBM leaves less flexibility for other DRAM products unless new capacity, better yields or process transitions offset the allocation.

Samsung began commercial HBM4 shipments in February 2026 and said in May that it had started shipping 12-layer HBM4E samples to major customers. Its 12-layer HBM4E is listed at 48GB, with 32GB and 64GB configurations also planned around customer requirements.

Samsung has also been accelerating future memory-fab investment, including its Pyeongtaek P5 expansion. New capacity matters because a growing HBM allocation creates larger tradeoffs when the underlying DRAM wafer base is constrained.

Consumer DDR5 and HBM serve different products, but they are connected upstream through DRAM manufacturing. A sustained increase in HBM's wafer share can reduce the amount of existing capacity available for conventional DRAM at the margin while AI accelerator demand remains strong.

For PC buyers, the evidence supports a narrower conclusion than saying AI will automatically make RAM more expensive. Samsung has quantified how large the HBM allocation could become in 2027 and acknowledged the shared-capacity tradeoff. Retail DDR5 supply will still depend on how quickly new capacity ramps and how conventional DRAM demand develops.