An inside look at SK Hynix $720 billion AI-fueled buildout that's taking over South Korea

SK Hynix, the leading maker of high-bandwidth memory, is pouring $720 billion into memory factories to meet AI demand. We got an exclusive first look.

An inside look at SK Hynix $720 billion AI-fueled buildout that's taking over South Korea

Exclusive look inside SK Hynix's $720 billion AI memory expansion

In South Korea, where a mountainous landscape makes building huge projects an incredible undertaking, the world's leading maker of high-bandwidth computer memory is investing $720 billion on what it says will be the largest network of memory factories in the world.

SK Hynix, whose market cap has jumped more than fivefold in the past year and now tops $1 trillion, is making a big bet that extreme demand from artificial intelligence is here to stay. Soaring prices exacerbated by capacity constraints are changing the fabric of the classic boom-and-bust industry, with AI chip giants like Nvidia willing to take on more risk as they scramble to secure supply.

The industry is reliant on just a few players. SK Hynix controlled 58% of the high-bandwidth memory, or HBM, market in the first quarter, followed by fellow South Korean giant Samsung and U.S.-based Micron, both at 21%, according to Counterpoint Research.

To help fund its ambitious buildout, SK Hynix, which trades on the Korea Exchange, reeled in $26.5 billion in July by listing shares on the Nasdaq, the most money ever raised by a foreign company listing on U.S. markets.

It's been a tough month for new investors. Since reaching a high of almost $195 on July 14, the U.S.-listed shares are down by about 21%, closing on Wednesday at $154.41. The drop has coincided with broader volatility in the AI trade, particularly chip stocks, and a decline in the Korean market, which gets about half its value from SK Hynix and Samsung.

But none of that will deter SK Hynix from pursuing its historic buildout. It's part of a larger plan unveiled by South Korean President Lee Jae Myung in June to double the country's memory production in five years, which includes construction of multiple new Samsung megafabs.

SK Hynix head of HBM design Myeong-jae Park shows CNBC's Katie Tarasov the Yongin cluster under construction in Yongin, South Korea, on July 21, 2026.

Jeniece Pettitt

CNBC got a behind-the-scenes peak into SK Hynix's expansion last month, with the first on-camera tour of the Yongin Cluster, as well as an inside look at memory cleanrooms in Cheongju, where the company is also undergoing a major infrastructure push.

'It's like a war'

Height is the most striking difference between U.S. chip fabs and the three SK Hynix sites we toured in South Korea. The first fab at the Yongin Cluster had walls built about halfway up the towering structure, which SK Hynix said will reach the height of a 50-story apartment building. It will have six cleanrooms connected over several floors, instead of sprawling out in a single story like at new fabs in Arizona from Taiwan Semiconductor Manufacturing Co. and Intel.

Made by stacking general-purpose dynamic random access memory, HBM is the highest bandwidth memory capable of rapid data access that helps AI processors execute tasks. Those stacks of HBM take up so much DRAM it's depleting the supply for consumer electronics, as AI giants sign contracts that are years longer than the typical short-term agreements for what was historically considered a low-margin commodity.

"It's like a war," said Chey Tae-won, chairman of SK Group, the controlling owner of SK Hynix, in an interview in Seoul. "Everybody wants to buy the memory chips. Without that, they cannot produce their AI computing and AI chips."

Tae-won told CNBC that "prices went up too fast" and that he hopes all this buildout will help. "I'm trying to do my best," he said.

SK Group chairman Chey Tae-won shows an HBM4 chip to CNBC's Katie Tarasov at the SK Hynix parent copmany's headquarters in Seoul, South Korea, on July 22, 2026.

Jeniece Pettitt

SK Hynix reported in July that it signed 10 long-term agreements with major customers. Nvidia 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.

Tae-won showed CNBC a wafer with a message on it from Nvidia CEO Jensen Huang that read, "Please make more." He said he meets occasionally with major industry leaders like Microsoft's Satya Nadella, Google's Sundar Pichai, and Meta's Mark Zuckerberg. Huang and AMD CEO Lisa Su came to South Korea in recent months for meetings about memory.

MS Hwang, research director at Counterpoint Research, said all the hotels near SK Hynix and Samsung's memory fab expansions are fully booked.

"If you name any company in Big Tech, they are all in Korea to sign a contract," he said.

SK Hynix also makes memory chips at three fabs in China, but is prohibited from selling its leading-edge HBM there because of U.S. export controls. China has its own emerging makers of the coveted memory needed for AI, including CXMT, which just had a blockbuster debut on the Shanghai market.

"It's a race, and now the counterparty of the race is China," Hwang said. "That's going to be a lot more dangerous than anything else in the world."

President Lee is now urging speed on the fab buildouts in South Korea, while the U.S. also scrambles to add capacity.

The first of four fabs is taking shape at SK Hynix's massive Yongin cluster in South Korea, shown to CNBC on July 21, 2026.

Richard Larkin

Race to build bigger and faster

Micron is spending $50 billion to build two huge new memory fabs in Boise, Idaho, with the first one scheduled to start making wafers in 2027. And in Clay, New York, it's building a $100 billion campus with up to four fabs.

SK Hynix is also building its first U.S. facility, where construction is set to conclude in 2028. The $4 billion fab in West Lafayette, Indiana, is for packaging — where memory chips will be stacked and wired together — rather than front-end manufacturing.

In addition to its Indiana plans, SK Hynix has existing U.S. operations. In 2020, it agreed to buy Intel's NAND business for $9 billion, eventually creating the subsidiary Solidigm, which is headquartered near Sacramento, California. In January, SK Hynix restructured that business and launched a $10 billion "AI Company" in the U.S., but there are no plans related to front-end manufacturing.

Regarding U.S. chip fab plans, Tae-won said SK Hynix has been studying possible sites for more than a month.

"I'm willing to do it, but the problem is that I really have to find the right place," he said.

In the meantime, Tae-won said memory is at a "turning point" with the next technological leap to custom HBM that will help insulate SK Hynix's massive investment from risk.

Custom HBM is now being co-designed with AI processors, a trend that's accelerating with next-generation HBM5.

SK Group Chairman Chey Tae-won on SK Hynix's massive memory expansion — full interview