SK Hynix HBM4E Samples Could Ship as Early as June; Samsung Already Ahead
nashnova research
SK Hynix (SK海力士) plans to ship HBM4E samples as early as June, well ahead of its original "second half" timeline; Samsung already delivered the industry's first 12-layer 48GB samples on May 29, triggering an early head-to-head certification race.
What is HBM4E, and why the rush on samples?
HBM4E is the seventh generation of High Bandwidth Memory — stacked memory chips purpose-built for AI processors, far faster and denser than conventional RAM. Mass production is planned for next year.
This means → customer validation and optimization must wrap up in the second half of this year. Whoever delivers samples first enters the testing queue first, and the fight for mass-production orders starts the moment samples ship.
In plain terms = a sample is not a showpiece — it is an exam ticket. The earlier it arrives, the more tuning time the customer gets, and the better the odds of locking in the order.
Samsung vs. SK Hynix — who is ahead?
Samsung announced on May 29 that it delivered the industry's first 12-layer, 48GB HBM4E samples, claiming the "world first" timestamp.
SK Hynix could follow as early as June — months ahead of the "second half" window disclosed on its Q1 earnings call. SK Group Chairman Chey Tae-won said: "Whenever the customer is ready, we are ready."
This means → the two companies' sample windows now overlap almost entirely, and second-half customer certification becomes a direct head-to-head contest.
What is distinctive about SK Hynix's technical approach?
Core dies use the 1c-nanometer process; the base die — the "foundation" chip at the bottom of the stack — is outsourced to TSMC on its 3-nanometer node.
SK Hynix currently has only one HBM4E customer. The market expects that customer to be Nvidia, with the chips destined for the next-generation AI accelerator Rubin Ultra, slated for release next year.
Nvidia CEO Jensen Huang visited the SK Hynix booth at Computex on June 2 and wrote on an HBM4E wafer: "Make more, please." This reflects just how directly downstream demand is pulling on HBM4E capacity.
Where does Micron stand — the third supplier?
Micron's HBM4 capacity ramp is on track, but HBM4E mass production is not planned until 2027 — a full generation behind Samsung and SK Hynix in timing.
Micron's HBM4E will use its sixth-generation 1γ process at the 10-nanometer class, marking Micron's first use of EUV lithography — extreme-ultraviolet light used to etch finer circuits — in volume production. The base die is also outsourced to TSMC.
This means → in the near term, HBM4E remains a Samsung-SK Hynix duopoly. Micron's later entry point limits it to competing for follow-on orders.
What does this mean for the broader HBM market?
An industry source noted that HBM competition is no longer just about performance — shipping timelines and certification progress are equally decisive.
TrendForce has observed that all three suppliers are shifting focus from yield competition toward pricing power and next-generation spec leadership, and expects long-term HBM contract prices to rise.
In plain terms = whoever clears customer validation first and locks in mass-production orders holds the stronger hand in next year's pricing negotiations. The finish line is not "who builds it first" but "who sells it first."
市场有风险，内容仅供研究参考，不构成投资建议。