Samsung to more than double HBM4 output next year
The clearest signal on 2027 memory supply is not a wafer number — it is a cleaning contract for the glass plates that hold HBM wafers together while they are ground thin.
Samsung Electronics is expected to more than double output of its HBM4 family next year — sixth-generation HBM4 and seventh-generation HBM4E — and the tell is in its materials purchasing, according to Seoul Economic Daily, citing semiconductor industry sources. Samsung will raise outsourced cleaning volume for glass carriers, the supports used in HBM production, to 50,000 sheets a month next year from 20,000 sheets a month this year. That requirement stood at 10,000 sheets a month as recently as last year, so the volume is set to have grown fivefold in two years.
A glass carrier is temporarily attached to the underside of an HBM DRAM wafer so the wafer can be ground thin and drilled without bending or cracking. Because HBM stacks multiple DRAM dies inside a fixed package height, thinning and warpage control get harder as layer counts rise, which is why the carrier is a leading indicator rather than a commodity: you cannot ship 12-layer and higher stacks without buying the plates that make them possible. Analysts quoted in the report note the caveats — carriers are reused after cleaning, and consumption varies with process loading and yield — but argue a 2.5-fold rise in related volume makes at least a twofold increase in HBM4 and HBM4E output highly likely. That is inference from a supplier signal, not a figure Samsung has published.
The scale the industry expects behind it: Samsung's HBM production is projected to grow nearly 40% to about 250,000 wafers a month next year from roughly 180,000 wafers a month this year, measured by average monthly wafer input. The more interesting number is mix. The HBM4 family is projected to rise from around 40% of shipments this year to about 80% next year as HBM4E ramps, which means the high-value product stops being the premium line and becomes the default. Samsung began mass-production shipments of HBM4 in February on 10-nanometer-class sixth-generation DRAM with a base die built on a 4-nanometer process, and shipped 12-layer HBM4E samples to customers including Nvidia in May.
Our take: the mix shift matters more than the doubling. Memory is already running tight enough that HBM4 is eating conventional DRAM capacity — supply has fallen below 10-day levels at Samsung and SK Hynix, per Tech Times — and if four-fifths of Samsung's HBM output is HBM4-generation silicon next year, then the product that carried a premium for being scarce is about to be the volume part, which changes what buyers are negotiating over and how much of the shortage is structural. We have tracked the arithmetic of this squeeze before: Micron's 3x wafer cost for HBM over DDR5 explains why capacity does not convert cleanly into output, and what HBM actually is is the background if the stack-count talk is new to you. The caution is that all of today's story is one materials order read by analysts. Samsung has not published capacity guidance, and Nvidia qualification for HBM4E is the gate that decides whether 80% is a plan or a hope.
What to watch: whether Samsung puts a capacity or mix number on the record at its next earnings call, and whether SK Hynix answers with its own HBM4E ramp.
Is an inference from a cleaning contract good enough evidence to plan around, or should memory forecasts wait for the companies to say it themselves? Tell us in the comments.
