In mountainous South Korea, where building large-scale projects is an incredible undertaking, the world’s leading manufacturer of high-bandwidth computer memory is investing $720 billion in what it claims will be the world’s largest network of memory factories.
SK Hynixwhose market capitalization has more than quintupled in the past year and now stands at more than $1 trillion, is making a big bet that the extreme demand for artificial intelligence will continue. Rising prices, exacerbated by capacity constraints, are changing the structure of the classic boom-bust industry, with AI chip giants like Nvidia willing to take on more risks to secure supply.
The industry depends on just a few players. SK Hynix controlled 58% of the high-bandwidth memory (HBM) market in the first quarter, followed by fellow Korean giants samsung and based in the US micronBoth are 21%, according to Counterpoint Research.
To fund its ambitious expansion, SK Hynix, which is listed on the Korea Exchange, listed its shares on the Nasdaq in July, raising $26.5 billion, the largest amount ever raised by a foreign company listing on the U.S. market.
It’s been a tough month for new investors. Since hitting a high of about $195 on July 14, the U.S.-listed stock has fallen about 21% to close Wednesday at $154.41. The decline coincided with increased volatility in AI trading, particularly semiconductor stocks, and a decline in the South Korean market, which derives about half its value from SK Hynix and Samsung.
However, none of this will prevent SK Hynix from pursuing historic construction. This is part of a larger plan announced by South Korean President Lee Jae-myung in June to double the country’s memory production over five years, and also includes the construction of multiple new megafabs made by Samsung.
Myung Jae Park, head of HBM design at SK Hynix, shows CNBC’s Katie Tarasoff the Yongin cluster under construction in Yongin, South Korea, on July 21, 2026.
jenice pettitt
Last month, CNBC went behind the scenes of SK Hynix’s expansion, taking the first camera tour of the Yongin cluster and touring the inside of a memory clean room in Cheongju, where the company is also undergoing major infrastructure improvements.
“It’s like a war.”
The most notable difference between the U.S. chip manufacturing plant and the three SK Hynix facilities we toured in South Korea is height. The first fab in the Yongin cluster will have a wall built halfway up the towering building, which SK Hynix says will be as tall as a 50-story apartment building. The six cleanrooms will be connected across multiple floors, rather than being spread out over one story like the new Arizona factory. Taiwan Semiconductor Manufacturing Co., Ltd. And Intel.
Made of stacked general-purpose dynamic random access memory, HBM is the highest bandwidth memory with high-speed data access to help AI processors perform tasks. These HBM stacks consume large amounts of DRAM, which is depleting the supply in consumer electronics. This is because the AI giants are signing contracts that are many years longer than typical short-term contracts, which have historically been considered low-margin products.
“This is like a war,” Choi Tae-won, chairman of SK Group, SK Hynix’s controlling shareholder, said in an interview in Seoul. “Everyone wants to buy memory chips. Without it, you can’t produce AI computing or AI chips.”
“Prices went up too fast,” Taewon told CNBC, adding that he hopes all this ramping up will help. “I want to do my best,” he said.
SK Group Chairman Choi Tae-won shows the HBM4 chip to CNBC’s Katie Tarasoff at the parent company headquarters of SK Hynix in Seoul, South Korea on July 22, 2026.
jenice pettitt
SK Hynix reported in July that it had signed 10 long-term contracts with major customers. Nvidia has just “secured a stable supply of HBM” and agreed to co-develop next-generation memory as part of a $500 billion deal with SK Group that includes building new data centers with SK Telecom by 2027.
Taewon showed CNBC a wafer with a message from NVIDIA CEO Jensen Hwang saying, “Please make more.” He said he occasionally meets with key industry leaders such as Microsoft’s Satya Nadella, Google’s Sundar Pichai and Meta’s Mark Zuckerberg. Huang and AMD CEO Lisa Su have traveled to South Korea in recent months for memory conferences.
MS Hwang, research director at Counterpoint Research, said all hotels near SK Hynix and Samsung’s memory factory expansion facilities are fully booked.
“You name any big tech company, and they all have contracts in South Korea,” he said.
SK Hynix also manufactures memory chips at three factories in China, but is prohibited from selling its cutting-edge HBM in China due to U.S. export regulations. There are emerging manufacturers in China that manufacture the memory needed for AI, such as CXMT, which just made a blockbuster debut in the Shanghai market.
“This is a race, and now the race is against China,” Huang said. “It would be far more dangerous than anything else in the world.”
President Lee is currently urging the acceleration of factory construction in South Korea, and the United States is also rushing to add production capacity.
The first of four fabs is taking shape at SK Hynix’s massive Yongin cluster in South Korea, as revealed on CNBC on July 21, 2026.
richard larkin
Compete to build bigger and faster
Micron is spending $50 billion to build two huge new memory fabs in Boise, Idaho, the first of which is expected to start producing wafers in 2027. And in Clay, New York, the company is building a $100 billion campus with up to four fabs.
SK Hynix is also building its first facility in the United States, with construction expected to be completed in 2028. The $4 billion factory in West Lafayette, Indiana, is for packaging (where memory chips are stacked and wired) rather than front-end manufacturing.
In addition to the Indiana plans, SK Hynix has existing U.S. operations. In 2020, it agreed to buy Intel’s NAND business for $9 billion, eventually forming a subsidiary called Solidigm, headquartered near Sacramento, California. SK Hynix reorganized its business in January to launch a $10 billion AI company in the United States, but there are no plans related to front-end manufacturing.
Regarding the U.S. chip manufacturing plan, Taewon said SK Hynix has been considering potential locations for more than a month.
“We’d be happy to do that, but the problem is we really have to find the right location,” he said.
Meanwhile, Taewon said memory is at a “tipping point” with the next technological leap into custom HBM that will help protect SK Hynix’s huge investment from risk.
Custom HBMs are now being co-designed with AI processors, and this trend is accelerating with the next generation HBM5.
“NVIDIA wants its own custom chip, Google wants its own customized HBM, so HBM is more than just a commodity,” Tae-won said. “It actually changes the status of the memory chip.”
WATCH: An exclusive look inside SK Hynix’s $720 billion bet on AI memory
