SK hynix said on 7 August 2026 that its board had approved about 54 trillion won for two new fabs. The company has led high-bandwidth memory since 2013, when it built the first one. Between those two dates sit five generations, each announced as a world first by the company itself. We went through its own record to separate what is documented from what is asserted.
Last verified: 7 August 2026. Every dated fact below comes from SK hynix’s own newsroom, which is both the point and the limit: the company is the source and the subject. The corporate history and the generation dates are fixed. The fab timetable, the Nvidia partnership scope and the demand forecast will move, and we have marked the two claims we could not check against anyone other than the company.
Editor’s opinion
R&D bought the lead. Capital has to hold it.
Our view is that the interesting thing about SK hynix’s HBM position is how unglamorous its origin is. The company shipped the first high-bandwidth memory in 2013, into a market with no volume application for it, and kept shipping a new generation every two to three years while nobody outside the industry was watching. The 2023 demand shock did not create that lead. It revealed one that already existed.
The counter-argument is that a cadence funded by one customer’s capital cycle is not a moat. Almost none of the 54 trillion won arrives soon. The company targets December 2028 and June 2029 for the two clean rooms, so the money does nothing for the HBM4 ramp it puts in the second half of 2026 and everything for a demand curve nobody can see yet. The forecast underneath that spending belongs to Omdia, which SK hynix quoted in its own investment announcement, and we have not read the Omdia report. Samsung and Micron are building as well. The view survives because the twelve-year record carries dates. It would not survive one generation in which SK hynix comes second.
Andrew Lee. This section argues a view rather than establishing a fact. It is not a forecast and not a recommendation, and the material it draws on is listed in the sources below.
The company that existed before the lead
The HBM lead did not come from a new company. SK hynix began in 1983 as Hyundai Electronics Industries, absorbed LG Semicon in 1999, and renamed itself Hynix Semiconductor in 2001. On 26 March 2012 it joined the SK Group, took its present name, and held a launch ceremony at which it declared the goal of becoming the world’s best semiconductor company.
Two things in that ceremony are worth separating. The declaration was an ambition. The position behind it was not: the company stated at the same event that it already ranked second among the world’s memory makers. So the starting point for everything that follows was three decades of DRAM manufacturing and a number two ranking, not a standing start.
That changes how you should read the HBM story. A firm fabricating memory since the 1980s already employs the process engineers, runs the packaging lines and holds the customer relationships that a newer entrant has to build from nothing. SK hynix designed the first HBM part inside that, not alongside it. So the question worth asking is not how the company got good at stacking DRAM, but why it chose to spend on a product that had no market for most of a decade.
Twelve years of world firsts, and who says so
The generation record is the substance of the claim, so the dates matter more than any single part. SK hynix says it developed the world’s first HBM in 2013 and announced a product built on through-silicon vias jointly with AMD in 2014. Five generations followed on a two-to-three year rhythm, each announced by the company as a world first.
If you are looking for the moat, it is the cadence rather than any one device. A competitor can copy a stack height; matching a release every thirty months across through-silicon vias, pseudo-channel operation, on-die error correction and a 2,048-terminal interface means rebuilding the same packaging line repeatedly without missing a customer window. The commercial payoff arrived in the third quarter of 2022, when the company began supplying HBM3 into Nvidia systems and described that as an industry first. That was roughly a year before high-bandwidth memory became a phrase used outside the industry.
| Date | Generation, as the company describes it |
|---|---|
| 2013 | First HBM developed. HBM1 ran about 1,600Mbps on a 2Gb die stacked four high |
| 2014 | Through-silicon via HBM announced jointly with AMD |
| 2018 | HBM2, adding pseudo-channel operation |
| 2020 | HBM2E in mass production: 16Gb die eight high, 3.6Gbps, 460GB per second, heat down 36 per cent |
| Oct 2021 | HBM3 announced. Sixteen channels, 6.4Gbps, first on-die error correction. Mass production June 2022 |
| Q3 2022 | HBM3 supplied into Nvidia systems |
| 12 Sep 2025 | HBM4 complete: 2,048 terminals, bandwidth doubled, power efficiency better by over 40 per cent, above 10Gbps against a JEDEC standard of 8. Ramp from the second half of 2026 |
What the 54 trillion won is actually buying
The 7 August 2026 board decision is the point at which the HBM story stops being an engineering story and becomes a capital one. SK hynix approved about 54 trillion won across two sites: 35.2 trillion for Yongin Y2 and 19.1 trillion for Cheongju M17, both intended as production bases for HBM and next-generation DRAM.
Read the dates before you read the amounts. The company targets December 2028 for M17’s clean room and June 2029 for Y2’s, so neither contributes to the HBM4 ramp it places in the second half of 2026. This is spending aimed at a demand curve two to three years out, and the stated basis for it is an Omdia forecast of 19 per cent compound annual growth in DRAM and NAND demand from 2025 to 2030.
The capital markets side moved first. SK hynix listed American depositary receipts on Nasdaq under SKHY on 10 July 2026, issuing 177.9 million ADSs at 149 dollars to raise 26.51 billion dollars, and on 25 July the SK Group and Nvidia announced a partnership above 500 billion dollars covering AI factories and next-generation memory, with a separate SK hynix agreement for joint development and long-term supply. What a US listing does and does not change for a Korean issuer we set out in our note on the Solidigm listing.
The Omdia number reaches you twice removed. SK hynix quoted it in the release announcing its own 54 trillion won of spending, and the Omdia report is paid material we have not read. Treat it as the company’s stated basis for the decision, not as an independent forecast.
What the record does not tell you
Every date and specification above is SK hynix describing SK hynix, and you should weigh it that way. The company is the right source for when a part shipped and what it does, and the wrong source for whether it came first or works best. It also states that it is the only firm to have developed and supplied every generation from HBM1 through HBM3E. We found no third-party certification of that, so we are not printing it as a fact.
The competitive frame is the second seam. Samsung and Micron are both building HBM capacity, and a thirty-month cadence only reads as a moat while the company stays in front of it. We measured the distance between the Korean makers and their nearest Chinese challenger separately, in the CXMT gap, and the same problem appears there: a technology lead is easy to assert and hard to date.
Timing, and what we have not priced
Clean room dates in 2028 and 2029 are announcements, not deliveries. Fab schedules in this industry slip, and a slip moves capacity into a different part of the cycle from the one the spending assumed. Nothing in the 7 August release commits the company to a date it cannot revise.
We have not tried to price any of this. A twelve-year development record tells you where a company has been, and the fab calendar tells you what it has committed to. Neither tells you what the shares are worth today.
Four claims in this article
Where each one stands, as at 7 August 2026
| Claim | Attributed to | Status |
|---|---|---|
| HBM generation dates and specifications | SK hynix newsroom, September 2022 and September 2025 | On the record, company-published |
| 54 trillion won across Yongin Y2 and Cheongju M17 | SK hynix board decision, 7 August 2026 | On the record, company-published |
| DRAM and NAND demand growing 19 per cent a year to 2030 | Omdia, quoted by SK hynix, 7 August 2026 | Cited, original not read |
| Only firm to supply every generation, HBM1 to HBM3E | SK hynix, company claim | Not confirmed — claim withheld |
| Largest ADR listing on record | SK hynix, company claim | Not confirmed — claim withheld |
The first two rows are the company reporting its own decisions and products, which is the correct source for both. The third is a paid forecast we could only read through the company quoting it. The last two are rankings, and a company cannot certify its own ranking, so we print neither.
The record is twelve years long, and the company wrote all of it. That is enough to date the lead. It is not enough to certify it.
Sources
Almost everything here comes from one place, and you should know that before you weigh it. SK hynix publishes its corporate history, its product specifications and its board decisions on its own newsroom, and for those facts it is the primary source. It is not a neutral source for priority or ranking claims, and we have separated those out above rather than passing them through. We link to a document only where we hold the exact address for it.
Corporate history and HBM development
- SK hynix Newsroom, SK hynix Declared its Taking off for ‘World’s Best Semiconductor Company’, 26 March 2012. Source for the 1983 founding as Hyundai Electronics Industries, the 1999 LG Semicon merger, the 2001 renaming to Hynix Semiconductor, the March 2012 entry into the SK Group, and the company’s own statement at that ceremony that it then ranked second among the world’s memory makers.
- SK hynix Newsroom, Continuing to Make HBM History: The Story of SK hynix’s HBM Development, 8 September 2022. Source for the 2013 first HBM, the 2014 through-silicon via announcement with AMD, HBM1 at about 1,600Mbps on a 2Gb four-high die, HBM2 in 2018, HBM2E mass production in 2020 at 3.6Gbps and 460GB per second with heat down 36 per cent, and HBM3 from the October 2021 announcement through June 2022 mass production and third-quarter 2022 supply into Nvidia systems.
- SK hynix Newsroom, SK hynix Completes World’s First HBM4 Development and Readies Mass Production, 12 September 2025. Source for the 2,048 input-output terminals, the doubled bandwidth, power efficiency better by more than 40 per cent, speeds above 10Gbps against a JEDEC standard of 8Gbps, and the second-half 2026 ramp.
The 2026 capital decisions
- SK hynix Newsroom, SK hynix Invests 54 Trillion Won in Yongin Y2 and Cheongju M17, 7 August 2026. Source for the board decision, the 35.2 trillion and 19.1 trillion won split, the December 2028 and June 2029 clean room targets, and the intended use as HBM and next-generation DRAM capacity.
- SK hynix Newsroom, SK hynix Lists ADRs on NASDAQ, 10 July 2026. Source for the SKHY ticker, the 177.9 million ADSs at 149 dollars and the 26.51 billion dollar raise. The release also calls it the largest ADR listing on record; we withhold that ranking above.
- SK hynix Newsroom, SK Group and NVIDIA Expand Strategic Partnership Across AI Factories and Next-Generation Memory, 25 July 2026. Source for the partnership above 500 billion dollars and for the separate SK hynix agreement covering joint development and long-term supply of next-generation AI memory.
Third-party material we could not open
- Omdia, forecast of 19 per cent compound annual growth in DRAM and NAND demand from 2025 to 2030, as quoted by SK hynix in its 7 August 2026 investment release. The Omdia report is paid material we have not read, so we attribute the figure twice and carry it as the company’s stated basis rather than as an independent forecast. No address held.
Corrections
None yet. If you can point to a third-party document that dates any of these priority claims, we will cite it and say what changed.
Disclosure. The author holds no position in SK hynix, Samsung Electronics, Micron, AMD or Nvidia as of the date of publication. This article is general information and analysis about one Korean issuer’s development record and capital spending. It is not investment advice, not a recommendation to buy or sell any security, and not a forecast of any price or index level. The section headed Editor’s opinion is comment: it argues a view, and we fence it off from the sourced material for that reason. We have not tailored anything here to your circumstances, and only a qualified adviser in your jurisdiction can do that. We are not registered as an investment adviser in any jurisdiction, and this site has no commercial relationship with any company, exchange or research provider named above. See the Editorial & Sourcing Policy and Disclaimer.