Nvidia Secures Korean AI Supply Chain as TSMC Capacity Strains Intensify
Nvidia CEO Jensen Huang announced deals with six South Korean conglomerates on June 8, securing memory supply and new data center customers across the region.
The anchor agreement was a multi-year technology partnership with SK Hynix, committing the chipmaker to developing advanced memory for global AI data centers.
"SK Hynix will continue to be Nvidia's largest memory partner," Huang said after meeting SK Group Chairman Chey Tae-won at the chipmaker's parent headquarters in Seoul.
Huang added that Nvidia already spends billions annually with SK Hynix, and that figure is set to grow substantially, per the same Reuters report.
He also warned that SK Hynix's plan to double memory wafer capacity by 2030 would still fall short of what AI demand will require.
SK Telecom agreed to build a gigawatt-scale AI cloud using Nvidia technology, with its first data center targeted for 2027.
Naver and conglomerate Doosan also signed on to deploy Nvidia technology in new AI data center builds, per the same report.
Huang met separately with LG Group's chairman to discuss AI for humanoid robots and future data center architecture, covering cooling and power delivery systems.
At Hyundai Motor Group's headquarters, Huang and Executive Chair Chung Euisun outlined a deeper alliance spanning autonomous mobility, robotics, and AI-powered manufacturing, per Bloomberg.
"No one is in a better position to create that than Hyundai," Huang told reporters, citing the automaker's scale in manufacturing and heavy industry.
Hyundai's roughly $5.9B Saemangeum initiative includes an AI data center, a robot manufacturing cluster, and a hydrogen plant, per the same Bloomberg report.
Huang called the site South Korea's "AI Valley" and said he would gladly join the project — so long as the barbecue pork remained excellent.
Intel's Opening
The Seoul deals landed on the same day that a separate constraint emerged. TSMC's manufacturing capacity is running out of room, and major AI chip designers are testing Intel as a backup, according to The Information.
Alphabet recently placed an order with Intel to manufacture more than 3M Tensor Processing Units in 2028. The order came after months of testing Intel's advanced packaging technology.
Nvidia has not yet placed an order but is running early trials of Intel's 18A manufacturing process, per the same report. The work is tied to Nvidia's Feynman GPU architecture, due in 2028.
TSMC CEO C.C. Wei told shareholders this week that global chip supply cannot keep pace with AI demand for years. That holds even as TSMC expands US manufacturing capacity.
Intel CFO David Zinsner said earlier this year that demand for Intel's advanced packaging was running in the billions of dollars annually. That figure was up from prior expectations in the hundreds of millions.
South Korea's benchmark Kospi index fell 8.3% on June 8 after strong US jobs data fueled bets on a Federal Reserve rate hike, per Reuters. SK Hynix shares fell 7.7%, and Samsung Electronics fell 10.2%.
Huang dismissed the rout, telling reporters that "the future of AI is very bright."
The combination of locked-in memory supply and a credible alternative to TSMC suggests Nvidia is quietly building redundancy into every layer of its supply chain.
