Samsung and SK Hynix in Talks With Seoul on a Second Chip Cluster as AI Orders Surge

Seoul is in active talks with Samsung Electronics and SK Hynix over a second semiconductor cluster, and construction once mapped for the 2040s is now being eyed for 2034 to 2035.

South Korea's government is negotiating the next round of large-scale semiconductor investment with its two memory giants, and a senior official says the buildout the country had been planning for late next decade now needs to start years sooner.

Kim Yong-beom, chief policy adviser to President Lee Jae-myung, told a panel discussion on Wednesday that Seoul is working with Samsung Electronics and SK Hynix on a new chip cluster, with a formal announcement to follow soon. According to Reuters, Kim described demand for AI chips as "exponential and explosive," and said that surge could require the two companies to bring forward construction of new facilities by more than ten years, targeting completion around 2034 to 2035.

"The question is how we will support the AI revolution," Kim said. He framed the next seven or eight years as a problem of physical capacity rather than corporate strategy, with the hardest part being where to put a second cluster large enough to matter.

A decade pulled forward

The acceleration traces back to a single product. High-bandwidth memory, the stacked DRAM that sits beside Nvidia and AMD accelerators, has become the binding constraint on training and running large AI models. Korea makes most of it.

Samsung and SK Hynix together control more than half of the global memory market, and by some industry counts the country supplies close to 90 percent of the world's HBM. The newest generation, HBM4, roughly doubles the interface width of its predecessor and pushes bandwidth past a terabyte and a half per stack, which is why accelerator makers will pay a steep premium for it.

The technology is also greedy. A given quantity of HBM eats around three times the wafer capacity of standard DDR5, so every wafer redirected toward AI memory is a wafer taken away from the chips that go into phones and PCs. That trade-off has spilled into the wider market, with some categories of memory rising more than 300 percent over a single three-month stretch and a consumer shortage that has reached graphics cards and ordinary RAM.

SK Hynix said in late 2025 that its memory output for all of 2026 was already sold out, much of it reserved by Nvidia. New fabrication lines take at least two years to bring online, so meeting demand that lands in 2034 means breaking ground long before. The global chip market is expected to approach 975 billion dollars in 2026, with memory the fastest-growing slice of it, according to industry trackers. Kim's remark put an official stamp on a conclusion the market had already reached.

The hunt for a second cluster site

Where the cluster lands has become a question of national policy.

Lee Jae-myung has made balanced regional development a signature priority and has pressed the big conglomerates to invest outside the greater Seoul corridor. Korean outlets have reported that Samsung and SK Hynix are weighing major commitments in the country's southwestern region, with figures that could reach into the hundreds of trillions of won. Those numbers run ahead of anything the presidential office has confirmed, and are better read as expectation than as signed deals. The Honam area, including possible advanced-packaging facilities in Gwangju, has been floated as a candidate.

The calendar is already filling up. Korea JoongAng Daily reported that SK Group Chairman Chey Tae-won is due in Gwangju on June 30 to announce a regional investment plan, while Samsung is expected to unveil what would be the country's largest AI data center in Asan, South Chungcheong, on July 2. President Lee held talks last week with Chey and SK Hynix chief executive Kwak Noh-jung, and was expected to meet Samsung Executive Chairman Lee Jae-yong this week to firm up the southwestern plans.

Later in the month, the presidential office plans a working-level session with conglomerate chief executives. Business leaders are expected to push back on the practical limits of building far from Seoul, raising the need for more power and industrial water while asking for tax incentives and lighter regulation.

The power problem behind the buildout

Money is the easy part of this equation. A modern fab and the AI data centers rising alongside it draw staggering amounts of electricity and water, and that is precisely what makes a remote site hard to deliver. When the presidential office sits down with conglomerate executives later this month, the agenda from the corporate side is expected to center on exactly these constraints.

Korea is layering data centers on top of the fab plans. A venture pairing OpenAI with Samsung and SK began moving toward construction earlier this year, part of a broader effort to keep AI computing capacity on Korean soil rather than renting it from hyperscalers abroad. Samsung's planned Asan facility, billed as the country's largest AI data center, fits the same logic.

SK Hynix rides HBM to the top of the market

No company illustrates the stakes better than SK Hynix.

On June 22 the chipmaker became South Korea's most valuable listed company, with reports putting its market capitalization at roughly 1.35 trillion dollars after its shares jumped 5.6 percent in a single session. The engine is HBM, where SK Hynix has built a lead that analysts expect to carry into the next product generation. Counterpoint Research has put its share of the HBM market above 60 percent at points in the cycle, and reports from Korea suggest Nvidia has earmarked close to 70 percent of its next-generation HBM4 orders, bound for the Vera Rubin platform, to the company.

Those chips carry extraordinary margins. By one reading of its first-quarter results, SK Hynix posted an operating margin near 72 percent, a figure that belongs to no ordinary memory business and reflects how much pricing power the AI build-out has handed to the few firms able to produce HBM at scale.

The company is already spending to keep up. It committed 15 billion dollars to new facilities in February and broke ground in April on a 19 trillion won advanced-packaging plant in Cheongju. A separate 120 trillion won cluster is planned in the Yongin area.

Samsung's parallel build

Samsung is moving on a comparable scale, though from a different spot in the AI memory race. It trails SK Hynix in HBM, having reached volume production of HBM4 later, but its share is climbing and analysts expect it to clear 30 percent of the market.

The company's domestic ambitions are large. Samsung has outlined spending of roughly 310 billion dollars over five years inside Korea across chip fabs, AI data center infrastructure, batteries, and displays, and it is assembling an advanced-packaging base around Cheonan and Asan. Its long-planned Yongin project envisions five advanced plants by 2042.

Abroad, Samsung is racing to narrow the gap with TSMC in contract manufacturing, where it holds about 8 percent of the foundry market against the Taiwanese leader's 70-plus percent. Its 37 billion dollar fab in Taylor, Texas, is preparing to start 2-nanometer production this year, anchored in part by a 16.5 billion dollar contract to build Tesla's AI6 chip.

A subsidy race Seoul cannot sit out

The investment talk arrives against a backdrop of competing national chip programs. The United States built its push around a 52.7 billion dollar package paired with a 25 percent investment tax credit. The European Union committed 43 billion euros, while Taiwan offers steep support for its chipmakers' land and construction costs. China remains the most generous backer of all.

Korea has answered with incentives that, on paper, rank among the most aggressive anywhere. The 2023 K-Chips Act lifted facility-investment tax credits into the 25 to 35 percent range and R&D credits as high as 50 percent. A 2024 package added roughly 19 billion dollars more, and Seoul has floated a 20.4 billion dollar public growth fund that blends state guarantees with private money to lower the risk of new projects.

The larger shift is a draft Special Semiconductor Act that would, for the first time, give legal backing to direct cash subsidies and to state support for the power and water that fabs consume in enormous volumes. The bill would also set up a Presidential Committee for Semiconductor Industry Competitiveness to run rolling five-year plans. One contentious clause, an exemption from the country's 52-hour workweek cap for high-earning chip researchers, was stripped out in committee after labor groups objected, and the issue remains unsettled.

Geography complicates the picture. Both companies still run sizeable operations in China, where US export controls limit how far they can upgrade advanced production, and that pressure is one more reason Seoul wants fresh capacity built at home. SK Hynix recently upgraded its DRAM fab in Wuxi, while Samsung's large memory operations in Xi'an sit under the same regulatory cloud.

There is also a people problem. Industry estimates point to a shortfall of more than 30,000 skilled workers across the coming decade, a gap that no amount of construction money closes on its own. The discarded workweek exemption was aimed partly at that bottleneck, and its removal leaves the labor question hanging over the entire program.

Why the world is watching

Korea's decision is being tracked well beyond Seoul. The country is one of only three places, alongside Taiwan and the United States, that can manufacture leading-edge memory at scale, which makes it a chokepoint in the AI hardware chain. American cloud providers and chip designers lean heavily on Korean memory, and Washington treats a US-aligned democracy with deep ties to its tech industry as a way to spread supply away from an overreliance on Taiwan.

That strategic weight gives Seoul a strong hand in dealings with allies, and it raises the stakes of every capacity decision the government signs off on.

Analysts also flag the other side of the trade. Memory is a cyclical business, and the same subsidy race pulling in investment across Korea, the United States, Europe and beyond carries a real risk of overbuilding if AI demand cools. Some forecasters expect HBM prices to soften once Samsung and Micron widen their supply, which would test the economics of fabs being approved today. For now, with capacity sold out and orders stacked into the next decade, that worry sits in the background.

Who captures the windfall

The chip boom has become a national economic story that reaches well beyond the two companies.

Korea is on track for double-digit nominal growth for the first time in more than twenty years, a pace driven almost entirely by the soaring profits of its chipmakers. That concentration worries some policymakers, Kim among them. He has cautioned that the windfall risks pooling in real estate rather than flowing into wages or productive investment, and has floated what he called a normalization of property taxation in response.

The distributional fight has already turned tense once. Samsung's largest union came within days of an extended walkout earlier this year before a government-mediated pay settlement pulled it back. As the AI cycle pulls fab construction forward, it is pulling those arguments forward too, over who benefits from the boom and over whether it strengthens the wider economy or simply enriches the firms sitting at its center.