Samsung expected to more than double HBM4 and HBM4E output next year

Samsung Electronics is expected to more than double output of its HBM4 family of high-bandwidth memory chips next year, according to a Seoul Economic Daily report dated September 20 that has since reached the Hacker News front page. The strongest signal comes from the supply chain rather than a product announcement: Samsung plans to raise its use of glass carriers, an essential material in HBM production, by roughly 2.5 times.

What the glass carrier numbers show

A glass carrier is a glass support temporarily bonded to the underside of an HBM DRAM wafer to stop it bending or cracking while the wafer is ground thin and drilled. That matters because HBM stacks multiple DRAM dies within a limited total thickness, and as stack counts rise, wafer thinning and warpage control become progressively harder.

Citing semiconductor industry sources, the paper reports that Samsung will increase outsourced cleaning volume for glass carriers from 20,000 sheets per month this year to 50,000 sheets per month next year. The ramp is steep by any measure: the requirement stood at 10,000 sheets per month as recently as last year, doubled this year, and is now set to grow another 2.5-fold.

The sources flag caveats. Carriers are reused after cleaning, and actual consumption varies with process loading methods and yield. Even so, industry analysts judge that a 2.5-fold increase in related volume makes it highly likely that HBM4 and HBM4E output will grow at least twofold from this year.

The generational shift under way

The HBM4 and HBM4E products Samsung is preparing to scale center on 12-layer and higher stacks. The company began mass-production shipments of HBM4 in February, built on 10-nanometer-class sixth-generation (1c) DRAM with a base die made on a 4-nanometer process. In May, it supplied 12-layer HBM4E samples to customers including Nvidia.

Overall HBM capacity is expanding alongside the generational change. The industry expects Samsung's HBM production scale, measured by average monthly wafer input, to grow nearly 40% next year to about 250,000 wafers, up from roughly 180,000 wafers this year.

The mix is shifting even faster than the total. The HBM4 family is projected to rise from around 40% of Samsung's HBM shipments this year to roughly 80% next year as HBM4E mass production ramps up. An industry official quoted in the report said Samsung appears to be putting HBM4, a high-value product, at the center of its expanded HBM production.

Why it matters

High-bandwidth memory is a gating component for AI accelerator hardware, and Nvidia is already among the customers that received Samsung's early HBM4E samples. If the carrier figures translate into shipments as analysts expect, the supply of leading-edge HBM available to accelerator makers will expand sharply just as the newest, highest-capacity generations arrive. For a component that has shaped how much AI hardware can actually ship, that loosening is significant on its own.

The mix shift carries a second message. Moving from roughly 40% to roughly 80% HBM4-family share means Samsung is not simply adding capacity but re-pointing most of its HBM lines at its most advanced product, prioritising margin and performance over volume of older generations.

One caveat is worth holding onto: the figures come from unnamed industry sources relayed in a single report, and Samsung has not publicly confirmed them. But the direction of travel — a supply of HBM4-class memory in 2027 that is at least double this year's — now looks difficult to miss.