Generated by Codex with GPT 5.6 Sol XHigh

SK Hynix has gone from near-collapse to the front of the artificial-intelligence boom. Once a struggling second-place memory-chip maker, it now leads the market for the high-bandwidth memory, or HBM, that feeds data to advanced AI processors. Its recovery shows how patient technical bets and a collaborative culture can let an underdog overtake incumbents. It also exposes the danger of expanding at the top of a notoriously cyclical market.

From rescue case to AI champion

The company nearly failed after the memory-chip downturn of the early 2000s. Sales fell by half, its market value sank to barely \$500m and a decade of restructuring followed. SK Group acquired it in 2012, when Samsung still dominated the industry and Hynix appeared destined to remain a distant runner-up.

That hierarchy has been overturned. SK Hynix briefly became South Korea’s most valuable company in June 2026, passed \$1trn in market capitalisation and raised a record \$26.5bn for a foreign company in its Nasdaq listing. Its quarterly sales were three times their level a year earlier, and management plans to double total capacity within five years.

The decisive product is HBM, which stacks memory chips vertically to move data much faster than conventional designs. SK Hynix began exploring the idea with AMD in 2008. The first commercial version, released in 2013, was too expensive, but the company kept developing it. That persistence mattered when AI systems suddenly created enormous demand for fast memory.

How the underdog moved ahead

Rivals helped create the opening. Samsung reduced its HBM team in 2019, sending experienced engineers to SK Hynix, while specialists also arrived from a delay-plagued Intel. SK Hynix then paired that influx of talent with a less hierarchical culture than Samsung’s. Employees were encouraged to share information, challenge decisions and turn unsuccessful experiments into formal case studies.

That approach produced practical manufacturing advances. One packaging technique fills the gaps between stacked chips with moulding material that helps dissipate heat. Improvements like this allowed SK Hynix to bring HBM3 to market ahead of Samsung in 2022 and become Nvidia’s sole supplier of cutting-edge memory. In an industry where capital requirements and technical complexity usually protect the largest incumbent, the former follower found a way to leapfrog it.

The boom may contain the next bust

Leadership creates a different test. Samsung and Micron are catching up, and both are expected to supply memory for Nvidia’s next server systems. All three producers are investing heavily just as Chinese manufacturers such as CXMT and YMTC are adding capacity. The familiar risk is that factories planned during a shortage begin production after demand has cooled, flooding the market and crushing prices.

SK Hynix has lived through that pattern before. When capital spending reached 40% of revenue in 2018, a server slowdown helped halve memory prices the following year; sales then fell by a third and operating profit collapsed by 87%. Management now promises to keep investment near one-third of revenue and says stricter long-term supply agreements make future demand more visible. Even so, Bernstein expects memory prices to peak in 2027 and SK Hynix’s sales to fall by 45% in 2028.

Politics adds another pressure. South Korea wants more semiconductor investment outside the Seoul region, while America is urging the company to build more locally. Those demands may support national industrial strategies without placing factories where talent and supplier networks are strongest.

SK Hynix’s rise is therefore both an industrial success story and a warning. The company won by investing early in an unfashionable technology, learning from failure and exploiting its rivals’ mistakes. Staying ahead will require the opposite kind of discipline: resisting the urge to mistake extraordinary demand for a permanent end to the memory cycle.