SK Hynix Adds Second-Largest Client; Nvidia Not Among Samsung's Top Five Customers
Nashnova编辑部
SK Hynix now has two customers each accounting for over 10% of revenue, while Nvidia is absent from Samsung's top-five client list — a gap that lays bare the two Korean memory giants' diverging positions in the AI accelerator supply chain.
What do SK Hynix's two "above-10%" customers tell us?
SK Hynix posted ₩131.9 trillion (≈$95.5 billion) in first-half revenue. Customer A contributed ₩17.6 trillion (13.35%); Customer B contributed ₩17.2 trillion (13.03%). Together they account for roughly 26.4% of total revenue.
Korea's Chosun Ilbo, citing industry sources, identified Customer A as Nvidia. SK Hynix did not name either customer in its filings.
This means → SK Hynix now has two core customers of comparable scale, not just one — and its customer concentration has actually fallen.
The top customer's revenue rose 60% — so why did its share nearly halve?
A year ago, only Customer A cleared the 10% disclosure threshold, contributing ₩10.9 trillion — 27.31% of that period's revenue. This year the same customer's revenue grew 61.7% to ₩17.6 trillion, yet its share dropped to 13.35%.
In plain terms = the biggest customer's business did not shrink. SK Hynix's overall revenue expanded even faster, stretching the denominator.
Customer B grew to a scale close to Customer A, triggering a separate disclosure. This reflects a broadening revenue base rather than a weakening top relationship.
The U.S. now accounts for 64% of revenue — where is the money coming from?
SK Hynix's U.S. revenue hit ₩84.6 trillion in the first half, up 203.8% year-on-year — roughly 64% of total revenue.
China rose from ₩7.4 trillion to ₩32.5 trillion; the rest of Asia contributed ₩11.4 trillion, Europe ₩2.8 trillion, and South Korea itself just ₩660.6 billion.
This means → SK Hynix's revenue center has shifted decisively to the U.S., driven most directly by AI accelerator demand.
Why doesn't Nvidia appear in Samsung's top five?
Samsung's top-five first-half customers were Alphabet, Amazon, Apple, Hong Kong Techtronics, and Supreme Electronics, together accounting for roughly 25% of Samsung's ₩305.4 trillion in consolidated revenue. Nvidia was not on the list.
This means → Nvidia's contribution to Samsung fell below that of the fifth-largest customer — implying a share below roughly 5% of consolidated revenue.
But the comparison has a structural limit: Samsung spans smartphones, consumer electronics, displays, foundry, and more. Even a large semiconductor buyer gets diluted in that consolidated pool. Put simply = it is not that Nvidia doesn't buy from Samsung — it is that Samsung's revenue base is too large and too diversified for any single customer to easily crack the top five.
HBM4 ramp — where is Samsung's catch-up window?
SK Hynix entered Nvidia's HBM — high-bandwidth memory, a stacked memory product built for AI chips — supply chain earlier, and is the primary supplier of HBM3E for Blackwell accelerators. Samsung entered the HBM3E chain later and at smaller scale.
Samsung began volume shipments of HBM4 and SOCAMM2 to Nvidia's Vera Rubin platform in Q1, expanded HBM4 sales in Q2, and sent HBM4E samples to major customers.
SK Hynix also started HBM4 shipments in Q2, plans full ramp in the second half, and delivered HBM4E samples on a 1c-nanometer process to an unnamed major customer.
What is the key second-half milestone to watch?
SK Hynix's first-half data already shows a pattern: the top customer's revenue grew over 60%, yet its revenue share nearly halved — overall expansion is dispersing customer concentration.
For Samsung, whether the Rubin-generation HBM4 ramp can meaningfully raise its share of Nvidia's supply chain is the most critical second-half checkpoint.
This reflects a competitive pivot for both Korean memory giants — from "who can make HBM" to "who can capture a bigger slice of the next generation."
Content is for reference only, not financial advice.