What SK Hynix makes for data centers
public · HQ Icheon, South Korea · skhynix.com · also known as: Solidigm
SK hynix makes DRAM, NAND flash, and high-bandwidth memory for AI servers, supplying HBM3E and ramping HBM4 for AI accelerators alongside enterprise SSDs sold through its Solidigm subsidiary. It has disclosed multi-year supply agreements with major AI customers including Nvidia.
Listing
Korea Exchange (KOSPI): 000660.KS· 1277000 KRW +4.42% · mkt cap 930.76T KRW
Snapshot 2026-10-25T10:41:17Z — nightly, not real-time.· Chart on TradingView ↗
Exposure
6 parts across 1 layer · 2 constrained
Q2 2026 preliminary results cite revenue growth driven by "high-value-added products, including HBM, DRAM for AI servers, and eSSD," with HBM4 mass shipments beginning in the quarter (news.skhynix.com, July 2026).
Parts, by layer
Layer 2 — Memory & storage
| Part | Role | Confidence | Competitors | Note |
|---|---|---|---|---|
| HBMconstrained | manufacturer | disclosed | BESI, Disco, Hanmi Semiconductor, Micron, Samsung Electronics | 62% of HBM revenue in Q2 2025, per Chosun Biz via Astute Group, 2025-09-24 -- a single-quarter figure, not a running average — leader |
| DRAMconstrained | manufacturer | disclosed | Micron, Nanya, Samsung Electronics, Winbond, CXMT | |
| Memory modules | manufacturer | disclosed | ADATA, Micron, Netlist, Penguin Solutions/SMART, Samsung Electronics, Kingston | |
| CXL controllers & memory expanders | manufacturer | disclosed | Astera Labs, Marvell, Microchip, Micron, Montage, Rambus | |
| NAND flash | manufacturer | disclosed | Kioxia, Micron, Samsung Electronics, SanDisk, YMTC | |
| Drives | manufacturer | disclosed | Kioxia, Micron, Phison, Samsung Electronics, SanDisk, Seagate |
Where SK Hynix sits in the stack
SK Hynix's placements in this catalog cover 6 parts, all in layer 2, memory & storage. 2 of them are tagged constrained, meaning a sourced shortage rather than a long lead time. Across these parts it overlaps with other companies most often on the lines below.
What these parts are
- HBMconstrained — High-bandwidth memory stacked directly on the same package as an AI accelerator, giving the GPU or ASIC very fast, close-range access to the memory it needs to stay fed.
- DRAMconstrained — Off-package system memory — DDR5 modules for host CPUs and LPDDR5X soldered near Nvidia's Grace CPU — that holds the working data a server's processors operate on.
- Memory modules — The physical DIMM (or module) that packages raw DRAM die, substrate, and management circuitry into the pluggable unit a server actually sockets.
- CXL controllers & memory expanders — Chips that let a server attach an extra pool of DRAM over a CXL link rather than only through the CPU's own memory channels, expanding usable capacity beyond what fits in local DIMM slots.
- NAND flash — Non-volatile flash memory die — the raw storage medium later packaged into SSDs — that holds data and model artifacts when the system is powered off.
- and 1 more in the table above.
Closest peers in this catalog
| Company | Shared parts | Including | |
|---|---|---|---|
| Micron | 6 | HBM, DRAM, Memory modules and 3 more | Compare → |
| Samsung Electronics | 6 | HBM, DRAM, Memory modules and 3 more | Compare → |
| Kioxia | 2 | NAND flash, Drives | |
| SanDisk | 2 | NAND flash, Drives | |
| Kingston | 2 | Memory modules, Drives | |
| ADATA | 1 | Memory modules |
Seen in real machines
SK Hynix is named in 6 of the rack teardowns and supercomputer breakdowns on this site. Each line below is a piece of that machine.
- Nvidia GB300 NVL72 · rack teardown
- HBM3E (288 GB per GPU)
- Grace CPU and LPDDR5X
- E1.S NVMe SSDs
- Google TPU7x (Ironwood) · rack teardown
- HBM3E (192 GB per chip, 8 stacks, 7.37 TB/s bandwidth)* (dominated Broadcom-Google HBM orders in H1 2025; secondary since)
- JUPITER · supercomputer
- GPU memory (96 GB HBM3 per Hopper GPU, 4 TB/s)* (reported as the primary HBM3 supplier for Nvidia's H100-generation GPUs; no source read confirms it for the GH200 units installed at JUPITER)
- Isambard-AI · supercomputer
- GPU memory (96 GB HBM3 per Hopper GPU, 384 GB per node)* (reported as the primary HBM3 supplier for Nvidia's H100-generation GPUs; no source read confirms it for the GH200 units installed at Bristol)
- Eagle (Microsoft Azure) · supercomputer
- GPU memory (80 GB HBM3 per H100, 640 GB per VM)* (reported as the primary HBM3 supplier for Nvidia's H100; no source read confirms it for the GPUs in Azure's ND H100 v5 servers)
- Eos · supercomputer
- GPU memory (80 GB HBM3 per H100, five stacks; 640 GB per DGX H100)* (reported as the primary HBM3 supplier for Nvidia's H100; no source read confirms it for the H100s in Eos)
From the blog
Latest feed
- Solidigm confirms plans to build a fab in the US, abandons IPO plans — Tom's Hardware, 2026-10-23SK hynix has decided to build a 3D NAND fabrication facility in the United States and has canceled its initial public offering plans for Solidigm.
- Solidigm to Discontinue Review of Potential IPO, Reaffirming Commitment to Long-Term Growth and Shareholder Value — newsroom:sk-hynix, 2026-10-22
- Samsung and SK Hynix are upgrading Chinese factories to produce advanced RAM chips as AI data center competition ramps up — Tom's Hardware, 2026-10-21Samsung and SK Hynix are reportedly upgrading their Chinese factories to produce advanced RAM chips to compete in the AI data center market against Chinese competitors.
- Samsung, SK Hynix reportedly upgrading China NAND fabs to fend off YMTC amid softening US stance — DigiTimes, 2026-10-21Samsung Electronics and SK Hynix are reportedly upgrading their NAND flash fabrication facilities in China to produce chips exceeding 200 layers to compete with YMTC, despite existing US restrictions.
- AMEC seeks South Korean foothold through MOCVD trials with SK Keyfoundry, Magnachip — DigiTimes, 2026-10-21AMEC is trying to enter the South Korean power semiconductor market by conducting MOCVD equipment trials with SK Keyfoundry and Magnachip, specifically noting that its competitors are currently unable to supply Samsung and SK Hynix's fabs.
- South Korean govt, SK Hynix launch KRW2 trillion loan guarantee program for chip suppliers — DigiTimes, 2026-10-20South Korean government entities launched a KRW2 trillion loan guarantee program to bolster semiconductor materials, parts, and equipment suppliers.
- Elon Musk points to China's 2028 DRAM supply surge to challenge Samsung, SK Hynix pricing power — DigiTimes, 2026-10-19Elon Musk suggested that a projected surge in China's DRAM supply by 2028 could challenge the pricing dominance held by Samsung and SK Hynix.
- SK Hynix advances chip inspection technology as 3D memory pushes conventional metrology to its limits — DigiTimes, 2026-10-19SK Hynix is advancing chip inspection technology to handle the challenges presented by 3D memory stacking.
- Micron develops robot memory as Samsung, SK Hynix target physical AI — DigiTimes, 2026-10-19Micron is developing robot memory technology, aiming to challenge competitors Samsung and SK Hynix in the physical AI market.
- AI memory demand lifts Samsung, SK Hynix 2027 outlook — DigiTimes, 2026-10-19Increased demand from AI workloads is boosting the 2027 outlook for memory manufacturers Samsung and SK Hynix.
Questions about SK Hynix
- What does SK Hynix make for AI data centers?
- This catalog places SK Hynix on 6 parts: HBM (manufacturer), DRAM (manufacturer), Memory modules (manufacturer), CXL controllers & memory expanders (manufacturer), NAND flash (manufacturer) and Drives (manufacturer).
- Is SK Hynix publicly traded?
- Yes — SK Hynix is listed on Korea Exchange (KOSPI) under 000660.KS. Prices shown on this page are a nightly snapshot, not real-time, and nothing here is investment advice.
- Where is SK Hynix based?
- Icheon, South Korea, per the company record in this catalog.
- Who else makes HBM for AI data centers?
- BESI, Disco, Hanmi Semiconductor, Micron and Samsung Electronics are also placed on HBM in this catalog.
- How well sourced are these placements?
- Every part→company placement carries a confidence tag. For SK Hynix the mix is 6 disclosed. "disclosed" means the company or a filing says so; "reported" comes from trade press or a reference-design list; "inferred" is a lead from the taxonomy with no citation yet. Treat reported and inferred as leads, not facts.
Answers are generated from this catalog's own rows — see sources & method.
Verification: enriched, last verified 2026-08-27. Sources: www.skhynix.com · news.skhynix.com.· Markdown · JSON