AI NewsWave 한국어

Korea AI News

SK hynix Tops 100 Trillion Won in Quarterly Revenue for the First Time, Sold Out Through 2029

By Chloe Park | Oct 28, 2026, 18:42 KST | Based on 19 Korean news reports

Key points

SK hynix announced its third-quarter 2026 results on Oct. 28, with quarterly revenue topping 100 trillion won for the first time. According to AI Times, consolidated revenue was 104.6155 trillion won, operating profit was 83.3338 trillion won and net profit was 39.7756 trillion won. Betanews reported that the operating margin approached 80%.

Digital Today explained that shipments of HBM4, the next-generation HBM, ramped up and prices for server DRAM and enterprise SSDs (eSSDs) rose, boosting sales of high-value data center products. EBN reported that the company plans about 100 trillion won in shareholder returns this year, and AI Times also noted a quarterly dividend of 20,000 won per share in its headline.

On the earnings call, SK hynix said that, counting long-term supply agreement (LTA) volumes it has already secured and demand under discussion with customers, its DRAM and HBM production through 2029 is effectively sold out. iNews24 reported that the company has signed LTAs with 14 customers, including large global cloud companies, and some customers have requested contracts longer than five years. IT Chosun cited the company as saying it has secured demand equal to its production over the next three years, and TechM reported that talks on further extending contracts are under way.

EBN and Chosun Biz reported that the company said HBM orders from customers are about twice the volume it can actually supply. On some customers reducing the number of stacked HBM layers, the company said its medium- to long-term demand outlook is unchanged. iNews24 reported that the company explained the move from 12 layers to 8 as a result of supply shortages and an example of demand overwhelming supply.

iNews24 reported that SK hynix said the key-value (KV) cache is growing rapidly with the spread of agentic AI and long contexts, and that storing all of it in HBM and DRAM runs into limits of capacity and cost. Chosun Biz reported that the company expects new demand for NAND flash as the approach of storing and reusing the large volumes of data generated during AI inference on SSDs spreads.

Related coverage continued around the earnings release. Citing Counterpoint, EBN reported that market attention is focused on quality certification and supply of the HBM4 to be used in Nvidia's next-generation AI accelerator Rubin. iNews24 reported that, citing a Bernstein analysis, Samsung Electronics may have overtaken SK hynix in third-quarter HBM sales. e-Focus (이포커스) reported that Anthropic's IPO filing names SK hynix, Samsung Electronics and Micron as HBM makers in its risk factors, and AI Times reported that SK hynix said on Oct. 23 it is halting its review of an IPO for Solidigm.

Glossary HBM (high-bandwidth memory) is memory that stacks DRAM in layers to widen data transfer bandwidth and is mainly used in AI accelerators. The KV cache is space where a large language model stores intermediate calculation results from context it has already processed.

Sources (in Korean)

Drafted with AI writing tools and published first; Chloe Park reviews it after publication. Translated from the Korean version.