Samsung Electronics confirmed on May 10, 2026 that it will restructure its leading-edge foundry operations around dedicated single-customer production lines, marking the most significant strategic shift at the unit since it overtook GlobalFoundries a decade ago. The pivot, announced by Foundry Business head Choi Si-young at Samsung's Hwaseong campus, follows months of speculation about persistent yield problems on the SF2 gate-all-around node and growing customer impatience with shared-fab variability.
What Samsung Actually Said
The company will dedicate two of its five planned SF2 modules at Pyeongtaek to single, exclusive customers under multi-year capacity reservations, with the remaining three operating on the conventional shared-multi-tenant model. The first dedicated line is widely reported to be reserved for a US AI accelerator customer, with industry sources naming Tenstorrent and a second unidentified hyperscaler as candidates for the second.
Yield on the SF2 node has been the elephant in Samsung's room. Internal estimates leaked through Korean media in March suggested SRAM bit-cell yield was running roughly 15 to 20 percentage points below TSMC's N2 at the equivalent ramp stage. The dedicated-line model is, in part, a yield-management strategy: by reserving entire modules for single customers, Samsung can tune process parameters tightly to that customer's design rules without compromising for second and third workloads.
Why It Matters
For two decades, the foundry business has run on a multi-customer, shared-fab model pioneered by TSMC. The economics depend on amortising tool capex across diverse workloads, which keeps unit costs low and utilisation high. Samsung's move pushes against that model. By dedicating modules, Samsung accepts lower asset utilisation in exchange for higher yield, faster cycle time and closer customer engagement on process tuning.
The dedicated model is closer to what Intel Foundry Services has attempted with its first external customers, and what Rapidus in Japan has explicitly built its early business around. It also resembles, distantly, the captive logic fab model that Samsung itself operated for its own Exynos business until the early 2010s.
Why Samsung Is Doing This Now
- Persistent SF2 yield gap versus TSMC N2 documented since late 2025
- Loss of a Qualcomm second-source slot to TSMC for 2027 designs
- Need to lock anchor customers ahead of SF1.4 ramp in 2028
- Pressure from group chairman Lee Jae-yong for tangible foundry returns
- Capacity overhang at Pyeongtaek with subdued shared-tenant demand
The Customer Side
For customers, the dedicated-line model has clear advantages. They get predictable cycle time, reserved capacity, and a much tighter feedback loop with the process engineering team. They get to influence process tuning in ways the shared model never permits. The trade-off is cost: Samsung will charge a meaningful premium for dedicated capacity, and customers must commit volume and pricing for multiple years.
For a hyperscaler training the next generation of AI accelerators, that trade-off may be worth it. For a smartphone chip vendor with seasonal demand, it is almost certainly not. Samsung's pivot is therefore as much a customer-segmentation move as a yield strategy: dedicated lines for hyperscaler and AI-accelerator workloads, shared lines for mobile and consumer.
What This Means for TSMC and Rapidus
TSMC remains the dominant force in leading-edge logic, with N2 production ramping smoothly at Hsinchu and growing volumes at Kaohsiung. Samsung's pivot does not change that immediate reality, but it could change the dynamic at the high end. If hyperscalers begin to value dedicated-line economics, TSMC may face pressure to offer something similar on N2 and N1.4, breaking its long-standing shared-fab discipline.
For Rapidus, Samsung's move is unexpectedly validating. The Hokkaido start-up has built its entire customer pitch around dedicated, single-customer shuttles, and the largest incumbent foundry just legitimised the model. Whether that translates into actual customer wins for Rapidus is a separate question.
Internal Reorganisation
The foundry restructuring includes meaningful organisational changes. Samsung is splitting the foundry sales force into two divisions: a Dedicated Capacity Sales group focused on hyperscaler and AI-accelerator accounts, and a Standard Capacity group covering mobile, automotive and consumer. The Dedicated Capacity group will report directly to Choi Si-young, reflecting its strategic priority.
Implications for the Korean Equipment Ecosystem
The dedicated-line strategy has knock-on effects across South Korea's broader semiconductor equipment and materials supply base. Companies like Hanmi Semiconductor, SFA Engineering and Wonik IPS, all of which supply Samsung's foundry operations, will need to adjust their support models to handle customer-specific process tuning at greater depth than the shared-fab model required. Industry observers expect at least two of the largest Korean equipment vendors to open on-site engineering offices at Pyeongtaek dedicated to specific customer programmes.
The materials side is more complex. EUV photoresist, advanced gas mixtures and ultra-pure chemicals have historically been sourced under shared-fab purchasing logic. The dedicated-line model raises the possibility of customer-specific material qualifications, which would meaningfully increase inventory complexity. Samsung's procurement team has been quietly working with key suppliers to model the resulting cost impact since late 2025.
What to Watch in the Next Six Months
Three signals will tell us whether the pivot is working. The naming of the first publicly announced dedicated-line customer, expected in the third quarter. The yield trajectory on the dedicated modules versus the shared modules; if dedicated yield converges quickly toward TSMC N2 norms, Samsung will have validated its bet. And, most quietly, the response from Apple and AMD, whose 2028 sourcing decisions for high-performance compute could decide whether Samsung Foundry is a credible second source or remains, structurally, the perpetual challenger.