SK Hynix expects memory supply shortages to persist through at least 2030, even as the South Korean memory manufacturer expands production capacity, CFO Kim Woo-hyun said on an October 28 conference call to announce third-quarter earnings2. The company has secured long-term agreements with 14 major buyers covering more than 50% of its output in 2027 and 20281.
Record quarter on AI-driven pricing
SK Hynix posted record third-quarter revenue of 104.6155 trillion won (approximately $78.1 billion), a 32% increase from the previous quarter3. Operating profit rose 38% quarter-over-quarter, and revenue exceeded 100 trillion won for the first time in the company's history.
DRAM average selling prices rose by a high-20% margin in the third quarter from the prior quarter. According to market researcher DRAMeXchange, the average price of DDR4 8Gb DRAM stood at $26 at the end of the third quarter, more than four times the $6.3 recorded a year earlier and up 23.8% from $21 at the end of the previous quarter. The average price of 128Gb NAND flash jumped eightfold year-over-year to $30.61.
Demand outlook tied to AI infrastructure
Kim said on the call that achievable memory demand next year would show bit growth in the mid-20% range for DRAM and the high-10% range for NAND flash. He added that supply would fall short of demand through 2030.
Citing data from global market research firms, Kim said capital expenditure by cloud service providers worldwide will rise more than 60% next year, while the AI data center infrastructure market will grow by an annual average of more than 40% through 2030.
ANALYSIS The shortage forecast and the scale of locked-in supply agreements reinforce the pricing power SK Hynix is already exercising, with DRAM prices quadrupling year-over-year and NAND prices rising eightfold.
The outlook aligns with constraints elsewhere in the semiconductor supply chain. ASML said on October 14 that AI-driven demand for its lithography equipment is so strong it will be unable to fully meet customer orders in 2027 or 2028, even after planned capacity expansions[3]. Separately, Nvidia Corp. has routed all NVHBM base-die production through TSMC's 3-nanometer process, pulling Samsung Electronics onto a rival's fabrication lines alongside SK Hynix and Micron Technology[2].
ANALYSIS With lithography tooling, advanced packaging, and now memory supply all constrained through the end of the decade, AI infrastructure buildouts face compounding bottlenecks that extend well beyond chip design.