SK Group Chairman Chey Tae-won called the surging memory prices caused by supply shortages "abnormal" and argued that prices should come down. He made the remarks during a press briefing at the 49th KCCI Jeju Forum held on Jeju Island, South Korea, on July 15, 2026. It may seem unusual for the head of a supply-side company that owns SK hynix to call for lower prices, but his reasoning is straightforward: if PC and smartphone prices keep rising, consumers will stop buying, and growth in the memory market itself will stall. What's more, he revealed that customer requests for AI-related demand in 2027 point to an increase of 60 to 100% compared with 2026. Faced with a choice between protecting short-term profits or expanding supply to protect the market, Chey made clear that he believes the latter is the right path.
A 60-100% Customer Ask, and a Widening Supply-Demand Gap for 2027
The 60-100% figure Chey cited is not a neutral forecast from a market research firm. It is a request-based figure he disclosed at the press briefing, reflecting how much AI-related semiconductors or memory customers say they will need in 2027. According to Maeil Business Newspaper and Hankyoreh, Chey indicated that AI-related semiconductors would account for more than half of total demand, and projected that overall memory demand would grow by at least 50-60% compared with 2026.
Manufacturers cannot expand supply at the same pace as demand. Chey explained that few manufacturers would be able to significantly increase supply in 2027, meaning the supply-demand gap would widen further compared with 2026. Given this, he rejected the strategy of restricting production to maintain high prices, arguing that even if profit margins dip somewhat, expanding supply and growing the market itself would be more beneficial in the long run.
Chey warned that if prices rise too far, it would invite new entrants into the market and increase the likelihood that governments around the world would pressure companies to expand supply. He also noted that a wide range of construction candidates were under consideration, mentioning South Korea's Yongin and Honam regions as well as the United States.
Behind a 93-98% Price Surge, the Device Market Shrinks
The word "abnormal" is backed by numbers. According to TrendForce, contract prices for commodity DRAM rose 93-98% quarter-over-quarter in the first quarter of 2026. DRAM industry revenue also swelled 81% year-over-year to $97 billion. TrendForce estimates that commodity DRAM prices will rise a further 58-63% in the second quarter, meaning rising revenue for suppliers and rising costs for device makers are happening simultaneously.
In smartphones, the impact of price increases has already begun feeding into product design. TrendForce estimates that the contract price for the mainstream configuration combining 8GB of memory with 256GB of storage rose roughly 200% year-over-year in the first quarter of 2026. Whereas memory previously accounted for about 10-15% of a device's bill of materials (BOM) cost, that share has now risen to 30-40%, forcing manufacturers to either pass costs on to consumers or cut back on storage capacity.
There are limits to how much cost can be passed on. TrendForce forecasts that global smartphone production in 2026 will fall 10% year-over-year to approximately 1.135 billion units, and that under a bearish scenario the decline could widen to more than 15%. For memory manufacturers, even if per-unit profit margins remain high in the near term, if rising prices trigger falling unit volumes, their customers' product markets will shrink. This is the demand destruction that Chey is wary of.
Even as Price Growth Slows, the Shortage Won't Disappear
For server DRAM contract prices in the third quarter of 2026, TrendForce forecasts a slower growth rate of 13-18% quarter-over-quarter. This is because multi-year long-term agreements (LTAs) have capped prices for some cloud service providers, and DRAM inventories have built up due to a shortage of server CPUs. However, this does not mean the supply shortage has been resolved. TrendForce estimates that Registered DIMM (RDIMM) supply in 2027 will grow only 15-20% year-over-year in bit terms, falling short of the growth in server CPU shipments.
The figures 60-100% and 15-20% refer to different things. The former is the AI-related demand request Chey received from customers, while the latter is TrendForce's estimate of RDIMM supply. Even so, both point in the same direction: growth in memory production for AI computing infrastructure is failing to keep pace with the growth in demand from major server memory needs. Even if price increases are capped for customers with LTAs, the burden of rising prices could shift onto customers without contracts or those making additional purchases.
Furthermore, high bandwidth memory (HBM) requires more wafers to produce the same amount of storage capacity than commodity DRAM does. The more production is shifted toward AI-related products, the harder it becomes to increase total supply without also expanding factory floor space and manufacturing equipment. The shift in the center of demand toward AI cannot be absorbed simply by changing the product mix at existing factories.
The 1,100 Trillion Won Strategy and the Time Lag Before Production Begins
SK hynix has announced a mid- to long-term strategy involving a total of 1,100 trillion won in phased investment across Yongin, Cheongju, and southwestern South Korea. The breakdown is 600 trillion won for Yongin, 100 trillion won for Cheongju, and 400 trillion won for the southwest. Chey's remarks this time do not represent a newly decided investment; the timing of execution will depend on market conditions and the company's investment principles. Curbing high prices hinges on how quickly the company can convert its already-announced plans into actual production capacity.
But factory construction cannot respond instantly to price signals. At Yongin, the completion target for Fab 4 has been moved up 12 years, from 2045 to 2033, but according to SK hynix, developing that same cluster took roughly nine years. At Cheongju, the company plans to invest 80 trillion won in the NAND fab M17 and 20 trillion won in advanced packaging facilities including P&T7. P&T7 is slated for completion by the end of 2027, while M17 is set to break ground in 2027 with operations targeted for the first half of 2029.
Factories require vast tracts of land and stable supplies of electricity and water. Beyond that, companies must also secure manufacturing equipment and skilled workers. Chey cited power infrastructure, cables, and even the raw materials for cables as the next bottlenecks after AI itself. It is a structure in which investments meant to expand memory supply run headlong into shortages in other infrastructure.
Given all this, it would be premature to read Chey's statement that "prices should come down" as a forecast of imminent price declines. Only once several things fall into place — how much actual supply allocation increases in 2027, how prices move for customers without LTAs, and whether P&T7 and M17 come online as scheduled — will it become possible to simultaneously sustain growth in AI memory and preserve the purchasing power of the PC and smartphone markets.