Samsung Expands Pyeongtaek Chip Megacluster as AI GPU Memory Demand Surges

Generative AI systems are powered by graphics processing units, or GPUs, and those processors increasingly rely on tightly integrated stacks of advanced memory chips packaged alongside them. As demand for that memory accelerates, Samsung is expanding what has become one of the world’s largest chipmaking footprints: a sprawling semiconductor manufacturing cluster in Pyeongtaek, South Korea.

Samsung already runs three fabrication plants at the site around the clock, and a fourth is now being brought online. Plans for a fifth factory were temporarily put on hold during the 2024 slump in memory-chip demand, but the pause did not last. With the boom in AI driving a sharp rebound, construction has restarted, turning Pyeongtaek—an agricultural city south of Seoul—into a front line of the global race to supply the components underpinning modern machine learning.

Just before midday on a Monday, the scale of that effort is visible as shifts break for lunch. Workers stream out of the fabs wearing black workwear and sky-blue vests, peeling off helmets labeled with their names and the facility number. Many move across the campus on scooters or motorcycles. The complex occupies roughly the area of New York City’s Central Park, sitting next to rice fields and surrounded by the infrastructure that has grown up to serve an enormous, long-term buildout.

The lunch crowd disperses into nearby eateries—buffets, diners, and restaurants serving beef stew—passing signs advertising short-term studio rentals. Some workers step into real-estate offices to ask about housing, sharing the same counters used by Samsung’s salaried engineers who are buying homes in the area.

Outside the fourth fab, Choi Hye-rin, a 35-year-old safety officer, describes her routine after arriving at dawn on a shuttle bus from Songtan, an older district nearby. Her team is building cable ducts, and her day is filled with inspections, equipment checks, chemical handling reviews, and written reports. She is one of hundreds of women working at the site; many, she notes, serve as spotters, walking ahead of heavy machinery to help prevent accidents in tight quarters.

Safety messaging is omnipresent. After lunch, excavators pass by steel frames that rise under tower cranes, while banners mounted high on girders urge workers not to rush even when schedules slip. A large LED board posts the winners of a safety-themed poetry contest. Hard hats and lockers carry pledges such as, “I will not work if it’s not safe.”

Yet the push is also driven by urgency. Samsung has forecast that memory demand will keep climbing through 2028. At the same time, the company has faced growing pressure from competitors: it has watched domestic rival SK Hynix gain an edge in high-bandwidth memory, the specialized products widely used in Nvidia’s AI accelerators. In the broader chip industry, Taiwan’s TSMC continues to dominate the foundry market, manufacturing chips designed by other firms.

For Samsung, the expansion in Pyeongtaek is a bid to regain momentum in memory—a category central to the performance and availability of today’s AI hardware. Despite summer heat and crowded work zones, Choi says she intends to stay. The pay is better than at many construction sites, she explains, and the work is comparatively manageable. She earns 140,000 won per day (about $97) and expects the AI-driven surge to keep projects active. “You just keep coming back,” she says.

Why it matters

AI’s growth is now constrained as much by physical supply chains as by software breakthroughs, and advanced memory is a key bottleneck. Samsung’s buildout in Pyeongtaek will shape how quickly—and how cheaply—companies can deploy next-generation AI infrastructure. The next major indicator to watch is whether Samsung can translate new capacity into leadership in high-bandwidth memory amid fierce competition.