In its third-quarter earnings report released on October 28, 2026, SK hynix said it expects memory demand to exceed supply through 2030. The company also said it plans to bring its M15X fab in South Korea to full capacity by the end of the year and to expand production of HBM4, the next generation of high-bandwidth memory (HBM) for AI. In other words, even as it adds capacity, SK hynix expects supply to stay tight for several years. To understand that outlook, it helps to separate when fabs are built and equipped from when products actually enter mass production and shipping. Comparing the Korean and U.S. capital investment plans announced in August with the latest earnings report shows capacity growing in stages, from increased output at existing sites in 2026 to mass production at a new site in late 2029.
Why SK hynix expects memory shortages through 2030
In its October 28 earnings announcement, SK hynix cited two constraints on expanding supply: advanced semiconductor manufacturing is becoming technically harder, and building new fabs takes a long time. Deciding to invest does not mean the products needed can be mass-produced right away. On the demand side, the company expects AI-related investment to keep growing.
SK hynix reasons that AI adoption will raise corporate productivity and create new revenue, and that those gains will flow into further AI investment. It also expects the spenders to broaden from today's large companies to a wider range of businesses and to national governments. In the short term, it sees growing AI use by companies and individuals pushing up memory demand; over the medium to long term, it expects the spread of autonomous driving, robotics and similar technologies to do so.
So the forecast does not say memory supply will fail to grow. It rests on the view that capacity takes time to expand while demand grows faster. Even with investment in new fabs, construction, equipment installation and establishing mass production all take time, and demand may keep rising in the meantime. Note, however, that "through 2030" is the period covered by the demand forecast; it does not mean the shortage will end in 2031.
Korean and U.S. expansion plans, by process and timing
When comparing SK hynix's investment plans, the key is the different manufacturing step each site handles. Chipmaking broadly divides into front-end processing, which forms circuits on silicon wafers, and back-end processing, which covers stacking, packaging and testing the finished chips. The new Indiana site will be a back-end facility: it will receive wafers made in South Korea and handle advanced packaging and testing for next-generation HBM. Indiana's capacity therefore cannot simply be added to the capacity of the front-end fabs in Korea.
For new fab plans, it is also necessary to distinguish between opening a cleanroom and starting mass production. A cleanroom is the controlled space in which semiconductor manufacturing equipment is installed and processes are run, but finishing one does not mean products can be mass-produced immediately. Mass production comes only after equipment is installed and manufacturing processes are brought up.
The table below summarizes each site's role and planned timing based on SK hynix's August 7 capital investment plan, its August 28 Indiana groundbreaking announcement and the October 28 earnings release. All entries are plans as of the announcement dates, not records of actual operations. The "next year" in the August materials is read as 2027, based on the 2026 announcement year.
| Site | Target / process | Construction start / ramp-up plan | First cleanroom opening | Mass production / operation plan | Source |
|---|---|---|---|---|---|
| M15X | HBM4 ramp-up | Ramp-up using existing facilities | Not stated | Full capacity by end of 2026 | October earnings |
| Y1 (Yongin Phase 1) | Product breakdown not stated | Not stated | February 2027 | Full-scale operation in 2027 | August investment plan / October earnings |
| Y2 | Front-end for next-generation DRAM (including HBM) | Construction starts July 2027 | June 2029 | Mass production date not disclosed | August investment plan |
| M17 | Front-end for NAND | Construction starts February 2027 | December 2028 | Mass production date not disclosed | August investment plan |
| Indiana | Back-end for next-generation HBM | Groundbreaking ceremony August 27, 2026 (U.S. time) | By October 2028 | Mass production begins in late 2029 | August groundbreaking announcement |
At the Indiana site, the plan is to open the cleanroom by October 2028 and begin mass production in the second half of 2029. Because the facility handles back-end processing of wafers made in Korea, adding back-end equipment alone will not raise HBM output if front-end capacity falls short. Actual supply capacity depends on how smoothly both front-end and back-end operations run.
For its new Korean fabs, SK hynix has also said it will expand cleanrooms and install manufacturing equipment in stages according to demand. Cleanroom opening dates for Y2 and M17 have been announced, but the company has not said when they will reach full capacity or how much actual shipments will grow. Investment periods run through October 2031 for Y2 and April 2031 for M17, but those dates should not be read as mass production start dates.
The scope of the facilities covered also matters when looking at investment amounts. The plans call for 35.2 trillion won for Y2 and 19.1 trillion won for M17, but Y2's investment includes not only manufacturing facilities but also office buildings, an integrated R&D building and related infrastructure. It is therefore not possible to assume the full amount goes to manufacturing equipment and estimate production capacity directly from the figures.
These plans show that output increases at existing sites come first, while new-fab capacity grows in stages over several years. Strengthening equipment that is already running and starting mass production at fabs yet to be built reach actual supply on very different timelines. That is why there is no contradiction in SK hynix pursuing large-scale investment while forecasting tight supply and demand through 2030. Still, these construction and operation plans alone cannot quantitatively prove that the overall memory shortage will last until 2030.
HBM4 ramp-up does not necessarily mean more PC memory
SK hynix expects both shipments and revenue of HBM4 to exceed those of HBM3E in the fourth quarter of 2026. That forecast reflects a shift in the product mix of its HBM business; it does not mean the whole memory market will move to HBM4 at the same time. Nor does a larger HBM4 ramp-up translate directly into more supply of general-purpose PC DRAM or SSDs.
In DRAM, the company plans to move to its "1c" process, the sixth generation of 10nm-class technology, and expand sales of DDR5 and LPDDR5 SOCAMM for AI servers. In NAND, it plans to raise the share of its 321-layer process in its South Korean production capacity to a majority in the fourth quarter. Here, "majority" refers to the share of NAND production capacity in Korea; it does not mean total NAND shipments will rise by 50% or more.
HBM, server DDR5, PC DRAM and NAND differ in their uses, manufacturing processes and demand trends. The HBM4 ramp-up plan therefore cannot be used to conclude that supply of PC DDR5 or consumer SSDs will grow at the same rate. Assessing market supply and demand requires checking production capacity and shipments product by product.
Long-term supply agreements should likewise be distinguished from actual manufacturing capacity. SK hynix says it has signed long-term supply agreements, typically five years, with 14 customers, securing contracted volumes equal to more than 50% of its total production capacity for 2027 and 2028. Contract prices are not fully fixed; they reflect market conditions within a price range agreed in advance.
Such contracts are an important input for understanding customers' medium- and long-term purchasing needs and for production planning. But securing contracted volumes does not in itself mean manufacturing equipment or shipping capacity has grown. Nor can the "more than 50% of total production capacity" figure be used to judge how large a shortage the global memory market faces, much less to conclude that PC memory supply will be cut in half.
What will decide the supply outlook through 2030: mass production progress
For 2027, SK hynix expects capital expenditure to rise sharply as Yongin Phase 1 reaches full-scale operation and production infrastructure is built out in and outside Korea. It says, however, that the specific scale and timing of investment will be adjusted for customer demand and competitiveness, as well as profitability and investment efficiency. The plan is not to execute all announced investments at once and convert them immediately into capacity. Because equipment will be added in stages to match demand, there will be a lag between investment progress and actual supply.
Judging whether the forecast that demand will exceed supply through 2030 holds up requires checking how much additional product shipment this investment actually produces. The earnings announcement and the August capital investment materials do not include figures that allow total market demand, product-by-product output increases and competitors' supply capacity to be compared on the same basis. Based on this information alone, it is therefore hard to estimate the scale of any future memory shortage quantitatively, or to predict when PC memory and SSD prices will fall.
The focus going forward is how far equipment installation and the establishment of mass production progress after the new fabs' cleanrooms open in 2028–2029. At the Indiana site in particular, what matters is whether mass production starts as planned in the second half of 2029 and whether the company can supply U.S. customers. At the same time, it will be necessary to see how much front-end capacity expands at the Korean fabs and how that shows up in shipments of HBM, general-purpose DRAM and NAND. The question is not construction plans but how closely actual shipments, product by product, keep pace with demand growth. That will be the most important indicator of how much SK hynix's large-scale investment eases the tight memory market.