SK hynix's 1c DRAM Share Goes From 13% to 34% This Year
The previous post looked at Samsung moving leading-edge logic capacity toward HBM4 base dies. This one looks at the other side of the same contest. SK hynix is accelerating a node transition on the DRAM side, and the timing of that transition explains a lot about how the two suppliers have chosen to compete.
KEY TAKEAWAYS
1. SK hynix's 1c DRAM share is estimated to move from around 10% in Q1 2026 to 13% in Q2 and 34% by Q4, reaching 35% in Q1 2027.
2. At that point 1c would pass 1b, which is projected at 33%, and become the company's primary node for the first time.
3. Samsung adopted 1c at HBM4. SK hynix stayed on 1b for HBM4 and introduces 1c at HBM4E. Same node, one generation apart.
Disclosed versus estimated
What the company said. On its July 29 earnings call SK hynix confirmed that supply of 1c-based products ramped during Q2. HBM4 entered volume shipment in Q2 with a fuller ramp planned for the second half, and HBM4E samples went to major customers in the first half. Management guided second-half bit growth above the first half, attributing that to HBM4 volume plus rising shipments of 1c-based commodity DRAM.
What is estimated. Chosun Biz reported on September 7 that SK hynix's 1c share runs roughly 10% in Q1 2026, 13% in Q2, 24% in Q3 and 34% in Q4, crossing 1b at 35% versus 33% in Q1 2027. The 1b share is described as peaking around 43% in Q2 2026 before declining. None of these figures come from a filing, and everything past Q2 is a forecast.
The same node, one generation apart
1c is the sixth-generation 10nm-class DRAM node. Which node the core dies are built on changes both the performance ceiling of an HBM stack and how hard it is to manufacture.
Samsung went to 1c at HBM4 and quotes 11.7 Gbps, about 46% above the 8 Gbps JEDEC standard. The company framed this as deliberately breaking its own precedent of using proven nodes for HBM.
SK hynix kept HBM4 on the proven 1b node and leaned on advanced MR-MUF packaging to hit its performance and cost targets instead. 1c arrives as an HBM core die only at HBM4E. On the earnings call the company described HBM4E as using a node chosen for maturity and manufacturing stability, which is a fair summary of the whole strategy.
So one supplier raised the specification first and took on the yield risk. The other secured the ramp first and deferred the specification step by a generation. Which was correct is not something this quarter can settle.
Why node transition is a cost line, not a slogan
Shrinking cell area raises the number of bits harvested from a single wafer. Same wafer starts, more output bits, lower cost per bit. That is the whole mechanism, and it matters more than usual right now.
Adding DRAM capacity means cleanrooms and tool installs, which is a multi-year commitment. Node transition adds supply inside lines that already exist. In a market where demand is running ahead of supply, conversion speed becomes an earnings variable rather than a technical footnote.
The three-way comparison is less tidy than the headline suggests. On industry estimates as of the end of Q2 2026, Micron sits around 19%, Samsung around 16% and SK hynix around 13% — meaning SK hynix is currently last. The same estimates put SK hynix at 34% and Samsung at 31% by Q4. No Q4 figure for Micron was given.
For now the scoreboard favors stability
Counterpoint Research and other firms forecast 2026 HBM4 share, across combined Nvidia, Google and AMD demand, at mid-50s for SK hynix, high-20s for Samsung and high teens for Micron.
If that holds, the HBM4 generation is being won by the supplier that shipped volume on a proven node rather than the one that shipped the faster part. A specification lead has not converted into a volume lead — at least not yet.
It is also too early to close the case. From HBM4E both suppliers are on 1c. At that point they compete on the same node, and what separates them is yield and customer qualification timing rather than node choice. The caution voiced in the Korean trade press — that conversion pace only means something if yield and qualification schedules keep up — applies to both companies equally.
What I actually watch
| Signal | Why it matters |
|---|---|
| Quarterly bit growth | The most direct evidence that a node transition is actually shipping. |
| DRAM operating margin | Cost improvement should persist in the margin separately from price effects. |
| HBM4E production timing | The first genuinely like-for-like comparison between the two suppliers. |
| Commodity DRAM supply | Whether 1c output goes only into HBM or also into server and SOCAMM parts. |
Two paths, side by side
| Samsung | SK hynix | |
|---|---|---|
| HBM4 core die | 1c DRAM | 1b DRAM |
| Differentiator | Foundry 4nm base die | Advanced MR-MUF packaging |
| First HBM use of 1c | HBM4 | HBM4E |
| Posture | Specification first, absorb ramp risk | Ramp stability first, defer specification |
Risks to this view
The quarterly node mix this post is built on is estimated, not disclosed. Everything past Q2 2026 has not happened.
Transition plans slip when yield does not cooperate. Planned mix and shipped mix are different numbers.
More bits per wafer does not help if early yield on the new node is low. The two effects offset.
HBM4 share forecasts are research firm views and vary between firms.
Closing
Samsung is moving capacity on the logic side. SK hynix is moving nodes on the memory side. Two routes to the same destination, and they converge at HBM4E.
Next post checks whether any of this shows up in hard numbers, with Micron's fiscal quarter results and TrendForce's Q4 contract price outlook both due in late September.
Sources: Chosun Biz, September 7, 2026 · SK hynix Q2 2026 results and earnings call, July 29, 2026 · Samsung Newsroom HBM4 mass production release, February 12, 2026 · Counterpoint Research and other HBM4 share forecasts as cited by Chosun Biz.
Comments
Post a Comment