Samsung Foundry's 4nm line is reportedly fully booked through 2027, marking a significant turning point for the long-struggling division.
The primary driver behind this is the explosive demand for HBM4 (High Bandwidth Memory), a critical component for AI accelerators. Unlike in the past, Samsung is now producing the logic 'base die' for its own HBM4 using its 4nm process. This strategic decision to 'internalize' production has created a powerful synergy between its memory and foundry businesses.
Let's trace the causal chain. First, the foundation was laid in February 2026 when Samsung announced the industry's first commercial shipment of HBM4, highlighting its DTCO (Design-Technology Co-Optimization) strategy. This approach, which integrates memory and logic design, was key to achieving the necessary quality and yield. Following this, reports in April confirmed that the 4nm yield had reached an impressive 80% and wafer prices for HBM4 base dies were soaring by 40-50%.
Second, this internal success was amplified by external market dynamics. Major customers like NVIDIA, with its upcoming 'Rubin' platform, and AMD have locked in their AI chip roadmaps for the latter half of 2026. This created a surge in pre-orders for HBM4 and, consequently, the 4nm base dies. At the same time, industry leader TSMC announced price hikes for its advanced processes, making Samsung a more attractive alternative and encouraging customers to diversify their supply chains.
Finally, these factors culminated in Samsung's official guidance during its Q1 2026 earnings call. The company stated its goal to run its advanced lines at full capacity in Q2 and explicitly linked foundry profit improvement to increased HBM4 base die production. This transformed the 'full booking' news from a one-off report into a confirmation of a fundamental business turnaround, suggesting the chronic deficit era is finally ending.
- Foundry: A business that manufactures semiconductor chips designed by other companies. It's like a 'factory for chips'.
- HBM (High Bandwidth Memory): A type of high-performance memory that stacks DRAM chips vertically to achieve faster data transfer speeds, essential for AI processing.
- Yield: The percentage of good, usable chips produced from a single silicon wafer. A higher yield means lower costs and higher profitability.
