SK Hynix Slows HBM4 Production Line Conversion, Opening a Catch-Up Window for Samsung
0xBroomberg
SK Hynix is delaying parts of its HBM4 line conversion to chase higher-margin commodity DRAM. This means Samsung now has a rare window to close the gap in HBM4 — and the memory market's share map faces a rewrite over the next two years.
Why is SK Hynix hitting the brakes at HBM's hottest moment?
The driver is not an HBM retreat — it is a structural margin reversal. Daishin Securities estimates commodity DRAM operating margins exceeded HBM's by more than 15 percentage points in Q1, and may approach the 90% theoretical ceiling by year-end.
This means → every fab line still running HBM instead of commodity DRAM is effectively leaving money on the table. Margins, not strategy, are redirecting capacity.
Nvidia's next-gen Rubin platform — the AI chip platform designed around HBM4 — has seen production expectations trimmed, reducing the urgency to convert at full speed. HBM already accounts for over 40% of SK Hynix revenue; the share is large enough that aggressive expansion is no longer necessary.
What made commodity DRAM suddenly so profitable?
SK Hynix's Q1 earnings show DRAM average selling prices rose by a mid-60% range quarter on quarter. In plain terms = the same memory chip costs roughly 60% more than it did last quarter.
The company also signed a three-year DDR5 supply agreement with Microsoft, locking in long-term revenue visibility. This reflects that commodity DRAM is not a short-term trade — large buyers treat it as a strategic resource worth a multi-year contract.
Management noted that Samsung is already earning substantial revenue from commodity DRAM. A rival's profitability reinforced the case for the shift.
How does Wall Street read this pivot?
Goldman Sachs argues that as long as SK Hynix holds above 50% share in HBM3 and HBM3E through 2026, earnings growth is supported.
Morgan Stanley went further: it raised SK Hynix earnings forecasts by 56% to 63%, citing an expected 62% rise in DRAM ASPs in 2026. This means → Morgan Stanley's thesis has shifted from "defend HBM share" to "the broader memory upcycle is the real value driver."
In plain terms = Wall Street is no longer pricing SK Hynix on a single HBM line — it is pricing the entire upcycle.
Has SK Hynix actually stalled on advanced HBM?
No. Reuters reports the company has shipped 12-layer HBM4E chip samples to key customers, with transfer speeds of 16 Gbps per pin and power efficiency gains of over 20% versus the prior generation.
Yonhap reports SK Hynix ordered thermocompression bonders — TC bonders, the core equipment for stacking multiple chip layers under heat and pressure — worth ₩44.2 billion (≈$28.7 million) from Hanmi Semiconductor, for delivery to its Cheongju M15X fab by early September.
This means → "slowdown" is a tempo adjustment, not a course change. Equipment orders and sample shipments both confirm the HBM capacity roadmap remains on track.
Can Samsung seize this window?
Counterpoint Research data shows SK Hynix held 57% of the HBM market in Q4 last year. Bernstein forecasts Samsung's HBM share will rise from 28% in 2025 to 35% in 2026 and 46% in 2027 — at which point SK Hynix would hold 37% and Micron 18%.
Samsung is expanding HBM4-related capacity at its Pyeongtaek P4 fab. Analyst Son In-jun at Eugene Investment & Securities expects Samsung to adopt the most aggressive HBM pricing in 2027, pushing both shipment volume and profitability simultaneously.
In plain terms = SK Hynix has voluntarily eased off the accelerator. Whether Samsung can convert that time gap into a lasting share advantage is the single biggest question in the memory industry over the next two years.
Content is for reference only, not financial advice.