Report: Samsung Has Allocated Half of Its HBM Capacity to HBM4
Alina Collins
Samsung has allocated half of its 150,000-wafer monthly HBM output to sixth-generation HBM4, booking roughly $1 billion in revenue within four months. It is a capacity-for-time gamble — betting on the next platform to claw back share lost in the last one.
How is the capacity split?
Of Samsung's 150,000 monthly HBM DRAM wafer starts, roughly 75,000 now go to HBM4.
The remaining half mainly produces HBM3E 12-layer stacks; the weaker-demand HBM3E 8-layer line has been temporarily halted and redirected to HBM4.
This means → Samsung is not "testing the waters" on HBM4 — it has committed half its HBM capacity outright.
Why the urgency?
In the previous generation, SK Hynix passed Nvidia's qualification first and captured most Blackwell orders. Samsung fell behind after a delayed certification.
This reflects a painful HBM3E lesson — rather than chase on an old battlefield, Samsung is sprinting to lead on the new one.
HBM4 targets Nvidia's next-generation Rubin platform. Whoever mass-produces first wins the orders; the window is the share.
Why does the ASIC wave favor Samsung?
Google, Amazon, and Microsoft are building custom AI accelerators to reduce dependence on Nvidia GPUs.
Starting with HBM4, the base die — the "foundation" chip at the bottom of an HBM stack — requires custom design, and advanced packaging matters far more.
Samsung operates memory, foundry, and advanced packaging under one roof. In plain terms = other vendors sell bricks; Samsung can build the whole house — a one-stop proposition that appeals to hyperscalers seeking custom solutions.
Why is SK Hynix in no rush?
Hynix has a strong customer base and shipping lead on HBM3E, and is expected to retain Nvidia's largest allocation in HBM4.
Nvidia CEO Jensen Huang said explicitly during his Korea visit this month that SK Hynix is his biggest supplier.
This means → Hynix's revenue will grow naturally as Rubin ramps; there is no need to pull capacity away from a still-profitable HBM3E line.
Nvidia CEO Jensen Huang said explicitly during his Korea visit this month that SK Hynix is his biggest supplier — Hynix has no need to rush HBM4 capacity; its revenue will grow naturally as Rubin ramps.
Semiconductor industry source
Anonymous industry insider
(Reported by Chosun Ilbo)
How will market share shift?
TrendForce and Bernstein project Samsung's HBM share rising from 27% last year to 37% this year; SK Hynix drops from 56% to 43%.
Since mass production began in February, Samsung's HBM4 has generated roughly $1 billion (≈ ₩1.54 trillion) in four months, with a full-year target of $10 billion.
Some analysts believe Samsung could overtake SK Hynix as early as next year to become the top HBM supplier. This reflects a market re-pricing of the race — not who ships the most today, but who is positioned earliest for the next generation.
Content is for reference only, not financial advice.