The $720 Billion Bet Behind South Korea’s Race To Power The AI Boom

South Korea is running out of room.

For SK Hynix, the world’s leading maker of high-bandwidth memory, that is becoming a very expensive problem to solve. The company is pouring $720 billion into what it says will become the world’s largest network of memory factories, betting that the extraordinary demand created by artificial intelligence is not another short-lived chip boom.

SK Hynix’s market value has surged more than fivefold over the past year, crossing $1 trillion, as demand for its memory chips has exploded. Prices have risen sharply as supply struggles to keep pace, changing the economics of an industry historically known for its brutal boom-and-bust cycles.

AI chipmakers such as Nvidia are now willing to take on more risk simply to lock in memory supplies.

The market itself is remarkably concentrated. SK Hynix controlled 58% of the global high-bandwidth memory, or HBM, market in the first quarter, according to Counterpoint Research. Samsung and US-based Micron followed with 21% each.

To help finance its expansion, SK Hynix raised $26.5 billion in July through a Nasdaq share listing, the largest-ever amount raised by a foreign company on US markets.

For investors who bought in recently, however, the ride has already been rough. The US-listed shares reached almost $195 on July 14 before falling about 21%, closing Wednesday at $154.41.

The decline has come alongside broader volatility in AI-related stocks and weakness in South Korea’s market, where SK Hynix and Samsung account for roughly half of the market’s value.

SK Hynix, however, is not slowing down.

The expansion is part of a much larger push by South Korea. President Lee Jae Myung unveiled plans in June to double the country’s memory production over five years, including several new Samsung megafabs.

‘It’s Like a War’

The biggest difference between SK Hynix’s South Korean fabs and the giant chip factories being built in the US is immediately visible.

Height.

At the Yongin Cluster, SK Hynix’s first fab is already rising into the skyline. Its walls have reached roughly halfway up the enormous structure, which the company says will eventually stand as tall as a 50-story apartment building.

Unlike many of the sprawling, single-level fabs being built by Taiwan Semiconductor Manufacturing Co. and Intel in Arizona, the Yongin facility will stack its operations vertically. Six cleanrooms will be connected across multiple floors.

The design reflects the extraordinary amount of memory now required to power AI.

HBM is made by stacking multiple layers of conventional dynamic random-access memory, or DRAM. The result is memory capable of moving data at extremely high speeds, making it critical to the AI processors used to train and run increasingly powerful models.

But those stacks require enormous amounts of DRAM.

As AI companies lock in supplies through contracts that stretch years into the future, memory once treated as a relatively low-margin commodity is suddenly one of the most strategically important components in the technology industry.

“It’s like a war,” SK Group Chairman Chey Tae-won stated. “Everybody wants to buy the memory chips. Without that, they cannot produce their AI computing and AI chips.”

Tae-won acknowledged that prices had risen too quickly and said he hoped the massive expansion would eventually ease the pressure.

“I’m trying to do my best,” he said.

SK Hynix said in July that it had signed 10 long-term agreements with major customers. Nvidia has also secured what it described as 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 plans to build new data centers with SK Telecom by 2027.

The desperation to secure memory has become almost visible – a wafer carrying a message from Nvidia CEO Jensen Huang: “Please make more.”

He said he periodically meets senior executives from some of the world’s largest technology companies, including Microsoft CEO Satya Nadella, Google CEO Sundar Pichai and Meta CEO Mark Zuckerberg. Huang and AMD CEO Lisa Su have also visited South Korea in recent months for discussions centred on memory.

The rush has spilled beyond the chip factories.

MS Hwang, research director at Counterpoint Research, said hotels around SK Hynix and Samsung’s expanding memory facilities are fully booked.

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

The China Problem

SK Hynix also operates three fabs in China. But US export controls prevent the company from selling its most advanced HBM there.

That has created another race.

China is developing its own HBM capabilities, with companies such as CXMT moving into a market that has traditionally been dominated by SK Hynix, Samsung and Micron. CXMT recently made 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.”

For South Korea, that makes speed more than a business priority. President Lee is pushing companies to accelerate new fab construction just as the US tries to build more domestic memory capacity.

Race To Build Bigger And Faster

Micron is investing $50 billion in two massive new memory fabs in Boise, Idaho. The first is scheduled to begin wafer production in 2027.

In Clay, New York, Micron is developing an even larger $100 billion campus that could eventually house four fabs.

SK Hynix is also expanding its US footprint.

Its first US manufacturing facility is being built in West Lafayette, Indiana, with construction expected to finish in 2028. The $4 billion facility will focus on packaging rather than front-end manufacturing. That means stacking and wiring memory chips together after they have been produced.

The company already has a substantial US presence.

In 2020, SK Hynix agreed to acquire Intel’s NAND business for $9 billion, eventually creating Solidigm, which is headquartered near Sacramento, California. In January, SK Hynix restructured that operation and launched a $10 billion “AI Company” in the US.

For now, however, that does not include front-end chip manufacturing.

Tae-won said SK Hynix has been studying potential locations for additional US fabs 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.

The bigger question is whether all this capacity will still be needed when the factories are finally ready. That is where SK Hynix believes the next generation of memory could change the equation.

From Commodity to Custom

Tae-won describes the memory industry as being at a “turning point.”

The next major shift is toward custom HBM, which is being designed alongside the AI processors it will ultimately power. That could make SK Hynix’s enormous investment less vulnerable to the traditional cycles that have defined the memory business.

The trend is already accelerating with next-generation HBM5.

“Nvidia wants their own custom chips and Google wants their own customized HBM, so it’s not just a commodity,” Tae-won said. “It actually changes the memory chip’s status.” 

That may be the most important bet inside SK Hynix’s $720 billion buildout. The company is not simply betting that AI will need more memory.

It is betting that memory itself is becoming a central part of the AI infrastructure — and that the days when it was treated as just another commodity chip are coming to an end.