SK hynix: The Complete Guide

The Q2 2026 figures are extraordinary enough to look like errors, and are not: revenue of KRW 79.3187 trillion, operating profit of KRW 60.5426 trillion "with an operating margin of 76%", and net profit of KRW 93.9226 trillion "with a net margin of 118%" 3, the same numbers filed with the SEC as provisional, unaudited K-IFRS figures 7. Net profit exceeding revenue is real and reported; neither the release nor the filing explains what produced it 3 7, and this page does not guess.

Two disclosures from the prospectus are worth more than the headline numbers. Customer concentration: "Our two largest customers represented 14.8% and 12.4%, respectively, of our total revenue in the first quarter of 2026 and our largest customer represented 23.9% of our total revenue in 2025" 6 — and the prospectus does not name them. And the listing created a second stream of SK hynix disclosure: the company now files with the US SEC as well as in Korea, including the Form 6-K carrying its quarterly figures 7 6.

SK hynix's own sources disagree about its most important date. The history page records "December 2013 — Developed the World's First TSV-based HBM" 1; the company's own HBM history article says SK hynix "jointly developed the world's first TSV (Through Silicon Via) HBM product with AMD in 2014" 10; and its 41st-anniversary piece dates the start of development to 2009 and says "four years later" the first-generation HBM arrived, pointing back at 2013 11. The honest reading is that development completed in 2013 and the first product was co-developed with AMD — a partner the history page's sole-first framing leaves out 1 10.

The mobile and automotive side is broader than the AI coverage suggests. The MWC 2026 lineup includes LPDDR6, UFS 4.1 at 1TB, uMCP 4.1 combining "16GB LPDDR5X + 512GB UFS", ZUFS 4.1, plus automotive LPDDR6 on the 1c node, automotive LPDDR5/5X, Auto UFS 3.1, Auto eMMC 5.1 and the PA101/PA201 automotive SSDs 15. LPDDR6 is listed as a product in SK hynix's own material, and for the automotive part the process is stated: automotive LPDDR6 and Auto/Robotics LPDDR5/5X are "both built on the 10-nanometer-class sixth-generation (1c) process technology" 15. What this check did not find in any SK hynix release is a speed figure or a shipping date for LPDDR6, so neither is stated here.

On NAND the official vocabulary differs from common usage. SK hynix's term is "321-high 4D NAND Flash", triple-level-cell, 1Tb 19 — not "321-layer", although its own Q2 2026 results page does use "321-layer products" 3. The enabling technique is what it calls "Three Plugs" process technology, which allows stacking over 300 layers, and the generation delivers a 59% productivity gain over its predecessor 19. Packaging is the other half of the story: SK hynix credits Advanced MR-MUF with "10% higher heat dissipation performance compared to the previous generation" and better warpage control on its 12-layer HBM3E 13.

The build-out is enormous and dated. M15X in Cheongju took KRW 5.3 trillion for construction against "more than 20 trillion won" long-term, for next-generation DRAM including HBM 9. In August 2026 SK hynix committed a further KRW 54 trillion: KRW 35.2 trillion for Yongin Y2, for "high-bandwidth memory (HBM) and other next-generation DRAM products", breaking ground July 2027 with cleanroom in June 2029; and KRW 19.1 trillion for Cheongju M17 for NAND, breaking ground February 2027 23. Indiana is "over $4 billion" for HBM advanced packaging, with cleanroom by October 2028 and mass production in H2 2029, alongside a Purdue University research testbed 22 24.

Two cautions. The Indiana figures moved: the July 2026 prospectus expected operations in H2 2028 6, where the August 2026 groundbreaking release says mass production H2 2029 22 — this page uses the later, more specific official figure and dates it. And China carries a named regulatory risk: "On August 29, 2025, the BIS announced that our designation as a VEU would be revoked effective December 31, 2025", replaced by "an annual approval mechanism", where "Failure to obtain the required licenses in a timely manner may materially impact our manufacturing operations in China" 6.

SK hynix's HBM4 deliberately exceeds the standard it implements. JEDEC's JESD270-4 sets 8 Gb/s per pin across a 2,048-bit interface, 32 channels per stack, up to 2 TB/s and a 64GB maximum 25; SK hynix's part runs at "over 10Gbps", which it states explicitly as being above the "JEDEC standard operating speed(8Gbps)", with power efficiency improved "more than 40%" and up to 69% better AI service performance 18. Note how the company frames the achievement: Justin Kim's quote is "We are unveiling the establishment of the world's first mass production system of HBM4" 18 — a claim about readiness to produce, not about having shipped first.

SK hynix's HBM4E figures are a straight generational delta on HBM4, stated against its own previous part rather than a competitor's: 48GB at up to 16 Gbps, power efficiency improved over 20%, and "improved heat resistance by 17 percent, compared to the preceding HBM4" 12. Notably, the HBM4E release makes no "industry first" claim at all 12, where the HBM4 and HBM3E releases both do 18 13 — a small but deliberate change in how the company is describing its own lead.

What SK hynix does publish, in place of prices, is the demand picture. Its own 2026 outlook post describes an "HBM-led memory supercycle" and republishes third-party forecasts — a $54.6 billion HBM market in 2026 from BofA, and custom-ASIC HBM demand growth from Goldman Sachs 27. Those are analyst numbers carried on an SK hynix page, not SK hynix forecasts, and this page attributes them to the houses that made them 27.

In July 2026 the relationship widened: "SK Group and NVIDIA today announced plans for a $500-billion-plus comprehensive partnership to establish AI infrastructure", within which "SK hynix is also entering into a long-term AI memory partnership with NVIDIA" that "allows NVIDIA to secure a stable supply of next-generation AI memory" 28 29. Read that figure precisely: the $500 billion is the SK Group–NVIDIA AI-infrastructure envelope, not the value of an SK hynix memory contract. No memory contract value is disclosed by either company 28.

Equally, SK hynix has never stated its share of NVIDIA's HBM4. Its own market-outlook page carries a 70% figure but attributes it to UBS: "UBS predicts that SK hynix will achieve approximately a 70% market share in the HBM4 market for NVIDIA's next-generation Rubin platform in 2026" 27. That is an analyst forecast republished by SK hynix, not an SK hynix claim, and this page does not convert it into one.

The July 2026 SK Group–NVIDIA announcement points at a customer relationship becoming an infrastructure one: it includes an SK Telecom "2-gigawatt-scale AI Cloud in Korea" built on NVIDIA systems and "powered by SK hynix HBM4", with a first AI factory planned for 2027 28. On the server side, SK hynix showed a "256GB 3DS RDIMM, which achieves the industry's largest capacity" at its own 2026 conference 14, and its framing of the strategy is broad: Jungdal Choi described the 321-high NAND as putting SK hynix "on track to advancing to the Full Stack AI Memory Provider" 19.

A third concerns the Intel NAND deal, and it is a gap in announcement rather than in substance. SK hynix marked the first closing in December 2021 with its own release, calling it "a paradigm shifting moment for SK hynix's NAND flash business to enter the global top tier level" and predicting Solidigm would be "the world's next big semiconductor company" 30. For the final closing, this check found no SK hynix release at all: the primary record is Intel's SEC filing, which states "On March 27, 2025, the second closing of the transaction occurred. The consideration received by Intel at the second closing of the transaction was approximately $1.9 billion, net of certain adjustments" 31. The deal was contracted at US$9 billion total — US$7 billion at first closing plus US$2 billion at final 21 — so the actual total came in slightly under the headline figure, and the only primary record of that is the counterparty's filing 31.

The disclosure gaps are narrower than Samsung's but real. SK hynix publishes no employee count in any source checked here; no exact SK square shareholding beyond the 20% statutory floor 6; no explanation of how Q2 2026 net profit exceeded revenue 3 7; and no LPDDR6 release giving specifications or timing despite listing the product 15. There is also no SK hynix announcement of the Solidigm final closing 31. Each is a case where the widely-quoted figure comes from somewhere other than the company.

For contrast, SK hynix's actual overseas manufacturing expansion is in the United States, not India: the West Lafayette, Indiana advanced packaging plant, over $4 billion, cleanroom by October 2028, mass production H2 2029, with around 7,000 direct and indirect jobs and a Purdue University research partnership 22 24. Ambassador Kang Kyung-wha described it at the groundbreaking as "the establishment of the first AI memory production facility and R&D center in the United States" 22.

On NAND the near-term plan is stated numerically: 321-high "products already represent the largest share of total production, and the company plans to expand this to approximately 50% of domestic production capacity by the end of the year" 3. The open questions are the ones the company has not answered: no HBM4E production date 12, no disclosed memory contract value inside the SK Group–NVIDIA partnership 28, and an annual US licence approval to obtain each year for the Chinese fabs 6.