Two new SK Hynix plants don't open their first cleanrooms — the dust controlled rooms where chips are patterned — until late 2028 and mid 2029. The $38B spending does not ease today's memory squeeze.
SK Hynix, the South Korean memory chipmaker, just approved 54.3 trillion won (about $38 billion) to build two new plants. The earliest of them won't open its first cleanroom until December 2028, and the second until June 2029. A cleanroom opening is still months away from shipping a single chip. The dollar figure is the trigger. The lead time is what sets the calendar.
The board's Aug. 7 decision splits the spending between two sites. The larger share, 35.2 trillion won (about $24.6 billion), goes to the Y2 fab in Yongin, a cluster SK hynix has been positioning as its next HBM and next-generation DRAM site. HBM, short for high-bandwidth memory, is the stacked memory used alongside AI processors. The rest, 19.1 trillion won (about $13.4 billion), funds the M17 facility in Cheongju, which is dedicated to NAND flash. Groundbreaking for M17 is set for February 2027; the Yongin site follows in July 2027. First cleanroom openings are December 2028 and June 2029, respectively.
A cleanroom is the dust-controlled, humidity-controlled room where wafers get patterned, etched, and layered. Coming online is when construction is finally done and equipment installation can begin. It precedes tool bring-up, qualification, and volume production by many more months. Reuters and the Korean 6-K filings walk the same construction-to-shipment timeline. The earliest any silicon from these plants could reach a customer is well into 2029, and most analysts put first volume later that year or in 2030.
Q2 results SK hynix released the same day put the demand pressure in numbers. Revenue hit 79.3 trillion won (about $55.5 billion), up 51% quarter-on-quarter and 257% year-on-year. Operating profit reached 60.5 trillion won (about $42.3 billion), on an operating margin of 76%. DRAM and NAND average selling prices both rose, and SK hynix framed the capex as a way to keep up with demand rather than pull forward supply. The company's own language: "This investment is a decision made to seize opportunities in line with the market's growth speed." That is consistent with a forward demand bet, and it is also consistent with the math the announcement implies: today's memory squeeze is not going to be relieved by a cleanroom that doesn't open until the back half of 2028.
Memory prices climbed through 2026. The Yahoo Finance analysis piece running alongside the capex news put DRAM and NAND contract prices both moving higher through the second quarter, on tight supply and AI-driven demand for HBM. SK hynix is the second-largest DRAM maker in the world and a leading HBM supplier. Its two new fabs are sized for the next memory cycle, not the current one. Samsung, the largest DRAM maker, and Micron, the third, are running parallel capacity additions on similar multi-year timelines, which is why one supplier's approval does not move the supply curve in 2026 or 2027. Today's memory pricing is set by what the three majors can ship from their existing footprint, because no new cleanroom from these two fabs will run tools before 2029.
HBM is sold largely on multi-quarter purchase commitments to a small set of AI accelerator makers, and the public documents do not name specific customer allocation. The capex release is silent on the question readers are most likely asking: does this $38B ease memory prices next year, or the year after? The honest read of the dates is that it eases neither. It positions SK hynix to compete for HBM market share roughly two years from now, in a market that is, by the company's own numbers, already growing faster than capacity. The near-term squeeze, in other words, is the company's reason for building, not a problem the building solves.
The board's vote was the trigger. The first chips from these two fabs will reach customers in 2029 at the earliest, and memory prices set in 2026 will not move on this announcement.