Infrastructure

38 Billion Dollars Buys Years of Pricier Memory Chips

SK Hynix's board approved $38.1 billion, 54 trillion won, for two new fabs in Yongin and Cheongju on 7 August 2026; the earliest new production does not arrive before 2028.

InfrastructureBy Servola Tech Desk2026-08-074 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

38 Billion Dollars Buys Years of Pricier Memory Chips

Key takeaways

SK Hynix's board approved 54 trillion won, 38.1 billion dollars, on 7 August 2026 to build two new fabs, Yongin Y2 for DRAM and HBM and Cheongju M17 for NAND, and neither one produces a single finished chip before 2028 at the earliest, so any relief in memory prices from this specific investment is years away, not quarters.
Yongin Y2 gets 35.2 trillion won and Cheongju M17 gets 19.1 trillion won; Cheongju breaks ground in February 2027 with its first cleanroom open in December 2028, while Yongin breaks ground in July 2027 with its first cleanroom open in June 2029, timelines that describe a company building for the end of the decade, not the current shortage.
Just nine days earlier, on 29 July 2026, SK Hynix reported 76 percent profit growth while holding capital spending steady; the following week its board approved 38.1 billion dollars in new fab commitments, a shift from restraint to acceleration that shows how fast AI memory demand is reshaping even a company's own stated spending plans.
SK Hynix itself frames the investment as a response to a structural shift, not a cycle, and expects both DRAM and NAND demand to keep growing at roughly 19 percent a year through 2030; for any European device maker or IT buyer, the practical read is to budget for elevated memory prices lasting years, not to wait out what SK Hynix does not believe is temporary.

The $38.1 billion decision

38.1 billion dollars. That is the figure SK Hynix's board approved on 7 August 2026 to build two new memory chip fabs in South Korea, 54 trillion won in total, confirmed in a statement on the company's own newsroom. Of that, 35.2 trillion won, about 24.9 billion dollars, goes to Y2, a second fab at the Yongin Semiconductor Cluster in Gyeonggi Province that will produce DRAM including high-bandwidth memory, HBM, the chip type most closely tied to AI accelerators. The remaining 19.1 trillion won, about 13.2 billion dollars, goes to M17, a new NAND flash fab at the company's existing Cheongju campus in North Chungcheong Province.

SK Hynix described the logic in its own words: the investment is "a decision made to seize opportunities in line with the market's growth speed", and the company said it wants to be recognised as "a key partner in AI infrastructure". That framing matters, because it treats the current run of AI-driven memory demand not as a spike to ride out but as a shift the company expects to keep shaping its business for years.

From holding capex to a 38 billion dollar pivot in nine days

On 29 July 2026, SK Hynix reported quarterly profit up 76 percent while holding its capital spending steady, a release that read, at the time, like a company choosing discipline over expansion even as AI demand ran hot. Nine days later, on 7 August, its own board approved 38.1 billion dollars in new fab commitments. Whatever the internal accounting bridges the two statements, the outward signal is stark: the company that said it was holding the line on spending a week earlier just made one of the largest single capacity commitments in its history.

The two fabs sit inside a larger plan SK Hynix first outlined in June 2026, worth 600 trillion won for the Yongin cluster over the long run and a further 100 trillion won to expand Cheongju. What changed on 7 August was not the existence of that ambition but its conversion into board-approved, dated, funded projects. For anyone tracking SK Hynix's spending discipline as a signal of where AI memory demand is really heading, the honest read is that the discipline held for exactly one earnings cycle.

Why the new capacity will not arrive before 2028, at the earliest

Cheongju's M17 breaks ground in February 2027 and is due to open its first cleanroom in December 2028, with the investment period running to April 2031. Yongin's Y2 breaks ground in July 2027 and is due to open its first cleanroom in June 2029, with its investment period running to October 2031. A cleanroom opening is not a shipping date; it is the point at which SK Hynix can begin installing and qualifying the tools that eventually produce chips at volume, a process that itself typically takes further quarters before meaningful output reaches customers.

Read plainly, that means the memory capacity these two fabs represent, DRAM and HBM from Yongin, enterprise-grade NAND and AI inference storage from Cheongju, does not meaningfully change global supply before 2029 at the earliest, and quite possibly later. SK Hynix itself is projecting close to 19 percent compound annual growth in both DRAM and NAND demand through 2030. A multi-year gap between when demand is already rising and when this specific new supply lands is, in effect, the company's own timeline for how long today's tight memory market should be expected to last.

What this means for European buyers budgeting memory costs

This is the clearest evidence yet that the AI-driven memory price pressure already visible in consumer and enterprise hardware pricing is a structural, multi-year supply response, not a temporary spike a European buyer can simply wait out. SK Hynix is one of the world's three dominant memory makers, and it has just told its own investors, in board-approved figures with fixed groundbreaking and cleanroom dates, that meaningful new capacity does not arrive until 2028 to 2029 at the earliest.

For any European device OEM, IT procurement team or hardware-dependent business owner, the practical planning consequence is straightforward: budget elevated DRAM and NAND pricing across the next two to three procurement cycles, not the next one or two quarters. A shortage a company itself is answering with multi-year fab construction, rather than a short production adjustment, is not the kind of shortage that resolves before the fabs it is funding actually open.

Frequently asked questions

Does SK Hynix's 38.1 billion dollar investment mean memory prices will start falling soon?

No. The earliest new production from either fab, Cheongju's M17 or Yongin's Y2, is tied to cleanroom openings in December 2028 and June 2029 respectively, with full investment periods running into 2031. SK Hynix itself is forecasting close to 19 percent annual growth in DRAM and NAND demand through 2030, so this specific investment is aimed at demand still to come, not at relieving the market that exists today.

Did SK Hynix reverse its position on capital spending in a single week?

The company reported holding capital spending steady alongside 76 percent profit growth on 29 July 2026, then had its board approve 38.1 billion dollars in new fab commitments on 7 August. Both fabs sit inside a broader 700 trillion won Yongin-and-Cheongju expansion SK Hynix first outlined in June 2026, so the August announcement formalises spending the company had already flagged rather than inventing it from nothing. Even so, the pace, from stated restraint to a board-approved multi-billion-dollar commitment in nine days, is itself a signal of how quickly AI memory demand is moving company decisions.

SK Hynix is not promising relief. It is building capacity for a decade of AI-driven demand it does not expect to ease, and the earliest of it does not open until 2028.

SK Hynixmemory chipsDRAMNANDHBMsemiconductor manufacturing

Three Firms Now Decide If Your Phone Stays Cheap

AI data centres are eating the memory chips that budget phones and laptops need. The sub-100-dollar phone is collapsing and prices hold high into 2027. What European buyers and fleet owners should plan for.

3 min read

Infrastructure

512 Gigabytes Now Sits Between HBM and the SSD

The first High Bandwidth Flash specification landed through the Open Compute Project: up to 512GB, three grades from 0.4 to 3.0 TB/s, on a UCIe interface. It is the industry routing around the HBM shortage.

5 min read

Decisions

The Codename Worked. The Trading Controls Did Not

An indictment unsealed on 24 July says two Volkswagen engineers traded on a joint venture known internally as Project Climb. The control that failed is a common one.