Nvidia's AI memory orders helped drive a sharp split between SK Hynix and Samsung Electronics in the first half of 2026. According to Chosun Biz, Nvidia became SK Hynix's biggest customer in the first half of 2026, while it did not rank among Samsung's top five revenue sources.
Samsung's first-half report listed its top five customers by revenue contribution as Alphabet, Amazon, Apple, Techtronics Industries, and Supreme Electronics. The chipmaker posted about KRW305.37 trillion (approx. US$221.3 billion) in revenue in the first half of 2026, with roughly KRW76 trillion coming from those five customers.
Based on that disclosure, Korean media estimated that Nvidia accounted for less than 5% of Samsung's total revenue. The top five customers together made up about 25% of Samsung's sales, reflecting a customer mix that leaned more heavily toward server memory, mobile DRAM, NAND, and general-purpose memory channels.
SK Hynix showed a far larger dependency on Nvidia-linked demand. One customer contributed about KRW17.61 trillion in first-half 2026 revenue, equal to 13.35% of total sales, and the industry widely believes that customer was Nvidia. That share was down from 27.3% in the first half of 2025, but revenue still increased 61.7% year-over-year from about KRW10.89 trillion.
South Korean industry sources said high-bandwidth memory 3E (HBM3E) remained the main product in the first half of 2026, helping explain the gap between the two memory makers' customer profiles. SK Hynix has supplied Nvidia since the early stages of the high-bandwidth memory market and continued to dominate HBM3E supply for Nvidia's Blackwell AI accelerator, while Samsung entered the HBM3E market later and is now looking to improve its position with HBM4.
Looking to the second half of 2026, shipments of Nvidia's Vera Rubin platform are set to begin, and HBM4 volumes are expected to rise, with both Samsung and SK Hynix expecting more revenue from Nvidia. Industry officials also said Alphabet, Amazon, and Microsoft are expanding in-house AI chip development and data-center investment, which should support memory demand.
Article edited by Ysi Chen
