SK Hynix (SKHY)

HBM memory leader (SK Hynix) — the dominant HBM supplier to NVIDIA

AI exposure 61% — derived from reported segment revenue, not asserted.

Where it sits

Silicon Layer → Memory & Storage ICs · figures rebased to Q2 2026, reported 2026-07-29

DRAM incl. HBM (high-bandwidth memory) — #1 HBM supplier for AI GPUs leads 2 businesses in this sub-tier that together are 100% of revenue, so Memory & Storage ICs is effectively the whole company rather than one leg of it.

Reported segments

• DRAM (incl. HBM) — 73% of revenue
• NAND flash — 27% of revenue

SK hynix (KRW 79.3T Q2 2026, +257% YoY, 76% operating margin). Q2 mix: DRAM including HBM 73% = primary B1 Memory & Storage ICs — the #1 HBM supplier, with HBM4 entering mass production in the quarter and DRAM ASPs up about 30%; NAND 27% (up from 21% as NAND prices rose in the mid-50% range), including Solidigm enterprise SSD. The DRAM share fell from 78% not because DRAM weakened but because NAND repriced faster.

How the 61% is built

• DRAM (incl. HBM) — 73% of revenue × 70% AI-driven (disclosed). HBM for AI accelerators plus data-centre DRAM dominate the unit; conventional mobile and PC DRAM is the remainder
• NAND flash — 27% of revenue × 35% AI-driven (disclosed). Enterprise SSD for AI storage tiers is the AI-linked part; client and mobile NAND is not

Investment read

Moat. HBM technology and capacity leadership — the primary HBM supplier to NVIDIA, with HBM4 mass shipments beginning in Q2 2026 and HBM4E samples completed in H1. A first-mover lead in TSV packaging yield that has to be re-won each generation, which is exactly what caps it below wide.

Bottleneck / pricing power. The HBM pricing-power leader, and now contracted rather than allocated quarter to quarter: long-term agreements cover roughly ten customers with multi-year locked volume, behind a record 76% operating margin in Q2 2026 — and management stated on the release date that major customers are still requesting more supply than SK Hynix can commit.

Role in the AI stack. The lead HBM supplier — the memory half of every AI accelerator.

Connected companies

• NVIDIA (NVDA) — SKHY supplies: HBM3E memory stacks
• Advanced Micro Devices (AMD) — SKHY supplies: HBM3E memory stacks for Instinct GPUs
• Rambus (RMBS) — SKHY is supplied by: comprehensive memory patent/IP license + DDR5/HBM interface IP
• Micron Technology (MU) — SKHY competes with: DRAM + HBM rivalry — both ramping HBM3E/HBM4 for AI accelerators
• Samsung Electronics (SSNLF) — SKHY competes with: HBM leadership fight — SK Hynix vs Samsung for NVDA qualification
• Onto Innovation (ONTO) — SKHY is supplied by: Dragonfly 2D inspection + 3D bump metrology for HBM; Onto's >$240M HBM volume agreement customer is not publicly named
• ASML Holding (ASML) — SKHY is supplied by: EUV/DUV for SK Hynix HBM/DRAM fabs
• Lam Research (LRCX) — SKHY is supplied by: etch/deposition for SK Hynix HBM/NAND
• Monolithic Power Systems (MPWR) — SKHY is supplied by: power management for memory/storage (MPS storage&computing)
• Apple (AAPL) — SKHY supplies: memory (DRAM/NAND) for Apple devices
• Alphabet (GOOGL) — SKHY supplies: HBM3E for Google TPUs (Ironwood/v7) plus server DRAM/NAND for Google AI infrastructure
• Meta Platforms (META) — SKHY supplies: server DRAM/eSSD procured for Meta AI data centers (plus SK hynix HBM inside Meta's NVIDIA fleet)
• Amazon (AMZN) — SKHY supplies: HBM/DRAM/NAND memory for AMZN AI data centers
• Oracle (ORCL) — SKHY supplies: HBM/DRAM/NAND memory for ORCL AI data centers

Others in Memory & Storage ICs

• Micron Technology (MU)
• Samsung Electronics (SSNLF)
• Sandisk (SNDK)
• Rambus (RMBS)

All 174 companies · What moved this quarter · How the map is built