Business & Economics
Nvidia Locks In SK Group for Memory and 1-GW Korean AI Cloud
During Jensen Huang’s 5–8 Jun 2026 Seoul trip, Nvidia converted SK hynix from a spot supplier into a multi-year HBM4 partner and, with sister firm SK Telecom, committed to build a gigawatt-scale AI cloud opening in 2027, signalling long-term, vertically-linked capacity planning rather than ad-hoc chip buys.
Focusing Facts
- Huang told reporters the SK hynix contract lasts “more than two years” with extension options and already represents “billions and billions of dollars” in annual purchases.
- SK Telecom and Nvidia will bring the first ‘AI factory’ on-line in 2027, designed for 1 GW of computing load—roughly ten times Korea’s largest current data-centre cluster.
- Industry estimates place SK hynix’s share of Nvidia’s HBM4 demand at 50-70 % across forthcoming Blackwell and Vera Rubin GPUs.
Context
The tie-up echoes IBM’s 1986 ten-year DRAM pact with Toshiba, when surging PC demand pushed OEMs to guarantee supply—only this time the scarce resource is 3-D-stacked HBM rather than planar DRAM. It also mirrors Ford’s 1920s River Rouge verticalisation: lock in raw inputs to keep the assembly line fed. The move fits a decade-long trend away from just-in-time semiconductors toward vaccine-style capacity reservations first seen in the 2020 COVID chip crunch, and it deepens the geographic concentration of AI infrastructure in an already tense Northeast Asian manufacturing corridor. Over a 100-year horizon, such alliances may either accelerate an AI-driven productivity wave or create new chokepoints where a handful of memory fabs wield OPEC-like leverage—especially if geopolitical shocks or export controls disrupt Korea’s ability to ship. Investors should note that press-release euphoria hides risk: a supplier providing up to 70 % of one company’s critical component is as much a single point of failure as a competitive moat.
Perspectives
Investor-focused financial media
e.g., 24/7 Wall St., Investing.com — Frame the Nvidia–SK hynix tie-up as a strategic masterstroke that will lock in years of profit by converting a memory shortage into a competitive moat for both companies. Coverage is slanted toward bullish takeaways that can excite traders—glossing over execution risks or over-reliance on a single supplier because the outlets cater to readers seeking stock tips.
International news-wire style outlets reprinting Reuters copy
e.g., bdnews24.com, CNA, Rappler, News.az, VnExpress — Report the series of South-Korean deals in a mainly descriptive tone, stressing Nvidia’s need to secure memory chips and noting related market moves such as the Kospi sell-off. By leaning heavily on corporate statements and Reuters prose, the pieces echo Nvidia’s messaging while giving limited space to independent analysts or dissenting voices, potentially underplaying competitive or regulatory hurdles.
Indian general-interest newspapers
e.g., The Hindu, Deccan Chronicle, Ommcom News — Highlight the gigantic scale of the planned AI data-centre build-out but underline missing details—such as undisclosed investment figures—and remind readers that chip shortages could persist well into the decade. With audiences less directly invested in Korean chipmakers, the papers stress uncertainties and long timelines, which can cast the announcement as aspirational rather than imminent, reflecting a regional distance from the hype cycle.
Like what you're reading?