1. SK Hynix memory shortage forecast: what the company said
South Korean chipmaker SK Hynix said on October 28, 2026, that it expects demand for memory chips to keep running ahead of supply through 2030. The statement came with its third-quarter results, in which preliminary revenue reached a record 104.6155 trillion won, the first time any quarter has topped 100 trillion won for the company. Chief Executive Kwak Noh-jung added that 2027 could prove the most severe year for memory supply the industry has seen.
The outlook matters well beyond Seoul. Memory chips sit inside nearly every computing device, and the fastest-growing category is the hardware used to train and run artificial intelligence systems. When the supply of those components is tight, the cost and timing of building data centers shift for every company that depends on them, from cloud providers to the developers of AI models.
2. A record quarter built on AI data center demand
The revenue figure of 104.6155 trillion won represents a milestone for a company that has spent decades inside a notoriously cyclical industry. Memory makers have historically swung between shortages and gluts, with prices rising sharply and then collapsing as new capacity arrived. A quarter above 100 trillion won suggests that the current upswing is of a different scale from earlier ones.
SK Hynix attributed the growth mainly to products serving AI data centers. High-bandwidth memory, usually shortened to HBM, led the way, followed by server DRAM and enterprise solid-state drives. HBM is built by stacking several DRAM dies vertically and placing the stack next to an AI accelerator, which lets the processor read data far faster than conventional memory allows. Server DRAM and enterprise SSDs fill other roles in the same facilities, holding working data and storing the large datasets that models learn from.
The company also reported a preliminary operating profit of 83.3338 trillion won for the period. Both figures were announced before the independent audit was complete, so they may be revised when the final numbers are published.
3. How memory cycles used to work
To see why the forecast stands out, it helps to recall how the memory business has behaved in the past. Memory chips are close to commodities: a DRAM module from one producer can usually be swapped for another's, so prices are set largely by the balance of supply and demand. When demand rises, manufacturers rush to build capacity, and when that capacity arrives all at once, prices fall and profits shrink. The pattern has repeated across several decades and has made the sector one of the more volatile corners of technology.
The present situation differs in one respect that the company emphasizes. The products in greatest demand, particularly HBM, are not interchangeable commodities in the same way. They are designed alongside specific AI accelerators, qualified by customers over months, and produced through demanding packaging steps. That makes it harder for new entrants or rivals to flood the market quickly, and it helps explain why buyers are willing to sign multi-year agreements instead of purchasing on the spot market.
None of this removes cyclical risk entirely. The company's projection assumes that spending on AI infrastructure continues to grow, and that assumption belongs to its customers rather than to the chipmaker. The forecast should therefore be read as a statement of what the producer sees in its order book today, not as a guarantee of conditions in 2030.
4. Why supply cannot catch up quickly
SK Hynix pointed to two structural reasons for expecting a prolonged imbalance. The first is the growing difficulty of manufacturing at the leading edge. Advanced memory requires more process steps, tighter tolerances and, in the case of HBM, complex packaging in which many thin dies must be aligned and bonded without defects. Each of those stages lowers the yield of usable products and slows the ramp of new output.
The second reason is the long lead time needed to build new fabrication plants. A new fab involves years of construction, followed by the installation and calibration of specialized equipment before commercial production can begin. A manufacturer that decides today to add capacity therefore cannot meet demand for a considerable time, which is why the company sees the gap persisting even as it expands.
The company has plans for new fabs in South Korea and the United States. Those facilities are part of its response to the demand picture, though the timing of their contribution is a separate matter, and the company's own assessment is that supply will remain short of demand through the end of the decade.
Investment in additional capacity also carries a financial dimension. A leading-edge fab costs a very large sum to build, and a manufacturer wants confidence that demand will still be there when it opens. Contracts that cover more than half of expected output for two years give the company a stronger basis for committing to construction in South Korea and the United States, because a large part of the revenue is effectively secured before the equipment is installed.
5. How do long-term agreements change the market?
Long-term supply agreements are now a central feature of how SK Hynix sells its products. The company has signed such agreements with 14 major customers, including large cloud service providers. The contracts typically run for five years, and their pricing follows market conditions within a range agreed with each customer, rather than fixing a single price for the whole term.
Together, those agreements already cover more than half of the company's expected production capacity for 2027 and 2028. For a business long exposed to abrupt price cycles, that is a significant degree of visibility. Customers commit to volumes well in advance, and the producer gains a base of demand that is less likely to evaporate if market sentiment turns.
The arrangement also shapes who gets access to scarce memory. Large cloud companies that can commit to multi-year purchases are in a position to secure supply first. Smaller buyers, and manufacturers of consumer devices and automobiles that rely on the same types of chips, may find less supply left to purchase in the open market, which tends to put upward pressure on prices there.
The practical consequence for customers is that planning horizons are lengthening. A cloud provider building a data center that opens in 2028 must now arrange its memory supply years beforehand, in the same way that it already reserves land, power and networking equipment. Memory, once an item bought close to the date of need, is becoming a strategic input that is negotiated far ahead of time.
6. Why 2027 could be the tightest year
Kwak identified 2027 as a possible peak in the gap between demand and supply. His reasoning, as relayed by the company, is that AI-related orders are arriving faster than new capacity can be brought online, and that the new fabs under development will not yet be contributing meaningful volume.
The forecast is the company's own projection and has not been independently verified. Memory demand depends on the investment plans of a small number of very large customers, and any change in their spending would affect the outlook. Still, the share of 2027 and 2028 output already spoken for by contract lends weight to the view that supply in those years will be limited.
7. The HBM4 transition and the M15X fab
Two operational milestones give a short-term read on how the company is executing. SK Hynix projects that HBM4, the newest generation of its high-bandwidth memory, will overtake its predecessor HBM3E in both revenue and shipment volume by the end of the fourth quarter of 2026. A swift handover to the newer product would indicate that customers are moving rapidly to the latest AI accelerators, and that the company's yields on the new part are sufficient to meet demand.
The second milestone concerns the M15X fab, which the company expects to reach full capacity by the end of 2026. Reaching that level on schedule would add output at a time when every additional wafer is already effectively sold. Delays at either point would likely tighten supply further, while smooth progress would show the company converting its backlog into revenue.
8. What it means for rivals, buyers and investors
SK Hynix competes mainly with Samsung Electronics and Micron Technology, both of which are also expanding advanced memory output and chasing the same AI customers. A shortage that extends to 2030 would, in principle, benefit all three producers, though each will be judged on its ability to deliver the newest products in volume and at acceptable quality.
For buyers, the message is that securing supply early carries a premium and that waiting carries risk. For investors, the combination of record results, the shift to HBM4 and contracted demand supports the argument that the company is better insulated from the swings that have defined the sector. The projections remain forecasts, however, and a sudden slowdown in AI spending would be the principal threat to them.
9. What to watch next
Three markers will show whether the forecast holds. The first is whether HBM4 passes HBM3E in revenue and volume by the end of the fourth quarter of 2026. The second is whether the M15X fab reaches full capacity on schedule by year-end. The third is how quickly the planned facilities in South Korea and the United States come online.
Final audited third-quarter figures and the company's next quarterly report will also indicate whether the contract coverage for 2027 and 2028 continues to rise. If it does, the idea of a memory shortage lasting to the end of the decade will look increasingly like a planning assumption for the whole technology industry rather than a single company's view.