Samples Stayed Home, Talent Went to Intel — The Week Korea's Memory Lead First Got Extracted

The most consequential Korean semi headline of the week was not SK Hynix's HBM4E 12-Hi samples, nor the KOSPI's first-ever break above 9,000. It was the news that Lee Seok-hee, former CEO of SK Hynix, has been hired as Intel Foundry's EVP for advanced packaging, scoped to scale EMIB-T (2.5D) and HBI hybrid bonding (3D). In the same week, President Donald Trump posted on Truth Social that Apple had agreed to produce chips with Intel in the US, with Intel's 18A-P node likely. And memory IP firm Netlist filed a fresh AI memory patent suit against Samsung Electronics and Google.

The three stories share one feature: Korea's memory leadership has become clear enough that multiple parties are now actively extracting value from it — all in the same seven days.

Two Different Lanes Out of Korea's Memory Franchise

This week, Korea's memory industry visibly bifurcated.

Lane one — samples and prices. SK Hynix began HBM4E 12-Hi (48GB) sample shipments exactly three weeks after Samsung's HBM4E announcement, using 6th-gen 1c (10nm-class) 32Gb DRAM core dies — ahead of its own 2H guidance. The cycle is moving with them: DDR5 16Gb spot printed $46.0 on June 19, Korea's May semi exports hit $29.4B (+154% YoY), and the Korean industry's 2026 global memory market forecast now stands at roughly KRW 1,500 trillion — 4.2x last year. Server DRAM has crossed 56% of the memory mix for the first time, eclipsing mobile and PC combined.

Lane two — talent, IP, and demand. Lee Seok-hee's move to Intel is not a one-off. The reported scope of his EVP role covers EMIB-T (2.5D interconnect) and HBI hybrid bonding (3D stacking) — both core back-end technologies that bind HBM to AI accelerators. For Intel to land merchant AI orders on 18A-P, the node alone is not enough; the packaging that connects node and HBM has to follow. The way TSMC won NVIDIA with CoWoS, Intel needs SK Hynix-grade HBM-logic integration know-how. That know-how walked into Intel's org chart this week.

On top of that, Trump's Apple-Intel announcement, even stripped of political theater, opens the door for the first Big Tech anchor to attach to Intel Foundry's 18A-P line. And Netlist's new HBM-adjacent suit hangs an IP claim over Samsung's memory revenue.

Intel's New Triangle — Node + Packaging + Demand

Intel Foundry has carried two visible weaknesses: (1) weaker HBM-to-logic back-end integration vs. TSMC, and (2) no Big Tech anchor. This is the week both gaps got partially filled.

Lee Seok-hee is a leader who has seen every HBM generation from HBM1 through HBM4 from inside SK Hynix. His decisions on Intel's EMIB-T roadmap will shape Intel Foundry's AI win-rate over the next 12-24 months. At the same time, if the Apple-Intel pact translates into actual wafer starts — Trump's announcement alone does not confirm this — Meta, Microsoft, and AWS's custom silicon foundry choices naturally diversify behind Apple.