Samsung is expanding its semiconductor manufacturing complex in Pyeongtaek, South Korea, into one of the most ambitious memory production sites in the world. The push comes as the artificial intelligence boom drives demand for GPUs—and, in turn, for the memory chips those GPUs depend on. Three fabs at the campus already run around the clock. A fourth is coming online, and a fifth is back under construction after a year when the economics looked grim.

A campus that never sleeps

The Pyeongtaek campus sits in Gyeonggi Province and has grown steadily since its first fab, known as P1, launched in 2017. The site now operates three active fabs, and the next two are where the company's ambitions are most visible.

Memory chips fall into two broad categories, and the AI buildout is straining both: DRAM provides the fast working memory that processors draw on, while NAND stores data over the long term. That split is now shaping how Samsung deploys its production lines.

At P4, Samsung is converting a production line from NAND to a dedicated DRAM line. The P4 DRAM line is targeting a production capacity of 100,000 to 120,000 wafers per month by Q1 2027. It will use Samsung's 1c process, the latest generation of the company's DRAM manufacturing technology.

The fab that was paused, then unpaused

P5 is the most telling piece of the story. Construction on the fifth fab was halted in 2024, when demand for memory chips dropped. The AI boom then sent demand soaring, and the work restarted. Construction on the P5 complex has since accelerated, with groundbreaking for the second line arriving six months ahead of schedule.

The pause-and-resume pattern reflects the cyclical nature of memory manufacturing, in which suppliers scale back during downturn and race to add capacity when demand returns—and because fabs take years to build, decisions made today shape supply well into the future.

Each P5 fab represents an investment of more than 60 trillion won. The fabs are designed with triple-deck cleanrooms to maximize output, and capacity targets run into the hundreds of thousands of wafers per month.

Shifting the production mix

Samsung is not only adding new capacity; it is also repurposing existing lines. The company is reallocating some NAND production capacity to DRAM at both its Pyeongtaek and Hwaseong facilities. The move is driven by a supply shortage in server DRAM, particularly DDR5, the current-generation memory standard in the data center servers that underpin AI and cloud services. Premiums on server DDR5 prices have reached 70%.

The financial results reflect the shift. Samsung's Q3 2026 operating profit surpassed 100 trillion won, a record quarterly figure bolstered by HBM (high-bandwidth memory) and DRAM sales. HBM is a premium form of DRAM built to move data at very high speeds alongside AI accelerators.

Beyond Pyeongtaek

Samsung is also backing a 360 trillion won investment plan for a cluster in Yongin, which will involve six new fabs. The first of those is projected for 2029—a timeline that illustrates how long leading-edge memory capacity takes to plan and build.

What comes next

The reallocation carries a potential knock-on effect in NAND: shifting lines toward DRAM could tighten storage memory supply at the margin.

Key milestones are now laid out on a calendar. These include whether P4 hits its 100,000 to 120,000 wafer monthly target by Q1 2027, whether the server DDR5 premium holds as new supply arrives, and whether Yongin's first fab stays on track for 2029.

Source: CryptoBriefing