SK hynix CEO Kwak Noh-Jung has forecast that 2027 will be the year of the most severe supply shortage in the history of the memory industry. In an exclusive interview reported by Reuters on July 10, 2026, he stated that while customer demand continues to grow, his company's production capacity has limits, and that the situation in which demand exceeds supply capacity will continue even beyond 2030. This is not a warning limited to HBM for AI accelerators. Overlaying SK hynix's earnings materials with its production expansion plans reveals a time lag: demand is spreading rapidly into general-purpose server DRAM and data center SSDs, but new manufacturing facilities cannot come online fast enough to keep pace.
After Selling Out 2026, a Shortage Awaits in 2027
CEO Kwak's forecast did not suddenly emerge from distant, pessimistic speculation. In its October 2025 earnings announcement, SK hynix explained that it had completed HBM supply discussions with major customers for 2026, and had also secured customer demand equivalent to its entire production output for the following year in both DRAM and NAND. The April 2026 earnings report likewise stated explicitly that an environment in which demand exceeds supply capacity continues.
In other words, 2027 is not the year the tightness begins. SK hynix sees the gap between demand and production capacity widening further after 2026, a year whose allocations are already filled. Kwak's phrase — "the worst year in the industry's history in terms of supply" — refers to the degree to which production increases cannot keep up with the absolute volume of demand. No shortage rate by product category has been disclosed, so this does not mean that all types of memory will be equally short.
The strength of supply and demand is also reflected in the company's financial results. In Q1 2026, SK hynix posted revenue of KRW 52.5763 trillion, operating profit of KRW 37.6103 trillion, and an operating margin of 72%. The company explained that sales of HBM, high-capacity server DRAM, and enterprise SSDs increased due to expanding AI infrastructure investment, and forecast that the pricing environment for both DRAM and NAND would remain favorable for the foreseeable future. The supply shortage is giving memory makers strong pricing power.
AI Inference Consumes Memory Beyond HBM
If we view the current tightness solely as a battle over HBM, we misjudge the breadth of demand. During large model training, the HBM stacked alongside GPUs drew the most attention. But as AI services shift toward inference, demand also grows for DRAM in the CPU servers that control processing, and for SSDs that store long context windows and search data. As of October 2025, SK hynix had already explained that AI computing loads were dispersing into general-purpose servers, spreading demand across an entire product lineup that includes high-performance DDR5 and enterprise SSDs.
By Q1 2026, this view had advanced further. The company noted that as AI agents repeatedly perform inference across diverse environments, the memory required increases across the board, from DRAM to NAND. When adding one AI server, it is not enough to prepare only the HBM used for accelerators — main memory on the CPU side and storage to hold models and search data are also required simultaneously.
On the manufacturing side, allocation to HBM is directly tied to the supply capacity of general-purpose DRAM. In its March 2026 earnings materials, Micron cited the diversion of a large share of wafers to HBM, HBM's high growth rate, and the slowing pace of bit-density gains achievable per wafer through miniaturization as constraints on DRAM supply. For NAND, the company explained that some cleanroom space is being converted for DRAM use, limiting available space. Expanding production of HBM and general-purpose DRAM competes for the same manufacturing resources.
Micron's outlook as of June 2026 points in the same direction. The company forecasts that supply and demand for both DRAM and NAND will remain tight beyond 2027, projecting industry-wide 2026 shipment growth (in bit terms) of roughly the low-to-mid 20% range for DRAM and about 20% for NAND. Even so, it judges that this will not be enough to fully satisfy demand. SK hynix is not alone in presenting an extreme view.
New Plants Open in 2027, But Supply Will Ramp Up Gradually
SK hynix is rushing to expand production. At M15X in Cheongju, equipment has already been installed to accelerate output of advanced DRAM. At the first fab of the Yongin Semiconductor Cluster, the company plans to invest approximately KRW 31 trillion in total to build six cleanrooms across two buildings. The first cleanroom's opening has been moved forward from the originally planned May 2027 to February of the same year.
However, the opening date of a cleanroom is not the date the supply shortage ends. Only after manufacturing equipment is brought in, production processes are ramped up, yields are raised, and customer qualification is completed will shipment volumes actually increase. Additional facility investment in Yongin is scheduled to continue through the end of 2030, and the plan does not call for all six cleanrooms to reach full capacity simultaneously. 2027 will be both the year new facilities begin operating and a year in which demand growth must be absorbed by existing plants and facilities still in the process of ramping up.
Expanding NAND supply will take even longer. In July 2026, SK hynix announced a plan to invest a total of KRW 100 trillion in Cheongju. This breaks down to approximately KRW 80 trillion for the new NAND plant M17, and approximately KRW 20 trillion for advanced packaging facilities including P&T7. P&T7 is scheduled for completion by the end of 2027, but construction of M17 will only begin in 2027, with operations targeted to start in the first half of 2029. M17 will not serve as an immediate remedy for the 2027 shortage.
Competitors' production expansion schedules line up along the same timeline. Micron plans to produce its first wafers at ID1 in Idaho by mid-2027, and at ID2 by the end of 2028. Meaningful shipment volumes from its existing plant in Tauyuan, Taiwan are also expected by mid-2027, and new advanced packaging capacity in Singapore is scheduled to begin contributing to HBM supply in the first half of 2027. New capacity from each company will be added gradually between 2027 and 2029. This is precisely why it is difficult to project a scenario in which supply surges in 2027 and the supply-demand balance reverses.
Long-Term Contracts Protect Supply, While Preparing for the Next Cycle
The supply shortage has also begun to change how memory is purchased. By June 2026, Micron had signed 16 strategic customer contracts, most of them five-year agreements spanning 2026 through 2030. The company states that remaining performance obligations based on minimum volumes and minimum prices total approximately $100 billion, while commitments such as deposits received from customers reach $22 billion. Rather than waiting to compete for procurement after plants are completed, customers are now locking in volumes years in advance through contracts, financially supporting manufacturers' capital investment.
Long-term contracts give customers visibility into future supply, and for manufacturers, they raise the certainty of recouping investment. Even so, the cyclical nature of the memory business itself has not disappeared. SK hynix posted an operating loss of KRW 7.73 trillion in 2023, with an operating margin of negative 24%. That figure reversed to 72% by Q1 2026. Should AI investment slow more than expected, there remains a risk of swinging into oversupply once the massive plants currently under construction ramp up further.
CEO Kwak's forecast extending beyond 2030 is a long-term corporate outlook that does not come attached to product-specific demand volumes or prices. It cannot be used to derive how much, or for how many years, consumer RAM or SSD prices will rise. What should be verified going forward is the pace at which wafer input and yields rise in Yongin from February 2027 onward, whether M17 can meet its planned start of operations in the first half of 2029, and whether the demand locked in through five-year contracts actually translates into real shipments. Overlaying the progress of these factors with demand growth and the pace of competitors' production expansion will make it possible to test SK hynix's forecast of a shortage extending beyond 2030.