Samsung Foundry Surges to 80% Capacity—How 4nm HBM4 and AI Demand Turn

Samsung Foundry Surges to 80% Capacity—How 4nm HBM4 and AI Demand Turned the Tide

After years of lagging behind TSMC, Samsung’s Pyeongtaek foundries are now running at near-full capacity, fueled by HBM4 production and AI chip orders. Here’s what’s driving the turnaround—and why it matters for the next generation of accelerators.

Samsung Foundry has quietly transformed from a struggling underdog into a high-demand manufacturer, with utilization rates now hitting 80%—a dramatic shift from just 50% a year ago. The reversal isn’t just about catching up; it’s about leading in a niche where competitors are playing catch-up themselves.

The pivot began with a focus on mature nodes—4nm, 5nm, and 7nm—that TSMC and others have long dominated. But Samsung’s strategy hinged on one critical advantage: HBM4 production, where its 4nm custom base die for high-bandwidth memory stacks gives it an edge in AI and data-center workloads. While rivals rely on older process nodes for HBM base dies, Samsung’s integration of logic layers directly onto the 4nm substrate allows for tighter, more efficient designs—ideal for ASICs and accelerators handling heavy AI computations.

This shift has translated into a surge in orders. External customers, from hyperscalers to AI startups, are now booking capacity at Samsung’s Pyeongtaek Campus P2 and P3 lines, which were barely scraping by last year. The demand isn’t just about legacy nodes; it’s also about leading-edge nodes, where Samsung is ramping up production for clients eyeing sub-3nm chips in 2027 and beyond.

The AI and Automotive Boost

Two high-profile deals are accelerating Samsung’s turnaround: Tesla’s AI5 ASIC and Apple’s image sensors. The AI5, a vehicle-grade chip for autonomous systems, is a cornerstone of Samsung’s profitability push, with production scaling up in Pyeongtaek. Meanwhile, Apple’s shift to Samsung for some of its sensor manufacturing has added another revenue stream, diversifying the foundry’s client base beyond traditional tech giants.

Key Upgrades and What They Mean

HBM4 with 4nm Custom Base Die: Unlike competitors using older nodes for HBM base dies, Samsung’s integration of logic onto 4nm allows for higher design density, reducing power consumption and improving performance for AI accelerators.
80% Foundry Utilization: A near-full capacity run across Pyeongtaek’s P2 and P3 lines, up from 50% last year and near-empty volumes in late 2024. This reflects a demand shift toward Samsung’s mature nodes for AI and data-center applications.
Tesla AI5 and Apple Sensor Contracts: High-profile wins that diversify revenue beyond traditional semiconductor clients, with the AI5 chip alone expected to contribute significantly to Q4 profitability.
Ramping Leading-Edge Nodes: While 3nm and sub-3nm production remains TSMC’s stronghold, Samsung is investing in external client demand for advanced nodes, positioning itself for long-term growth in AI and HPC markets.

The implications are clear: Samsung is no longer chasing TSMC’s lead. Instead, it’s carving out a specialized niche in high-bandwidth memory and AI-optimized silicon, where its 4nm HBM4 and custom logic layers give it a competitive edge. For customers, this means faster access to high-performance memory stacks—and for Samsung, it means a foundry business that’s finally turning a profit.

With TSMC still dominating sub-3nm, Samsung’s focus on mature but high-demand nodes ensures it won’t be left behind. The question now isn’t whether Samsung can compete—it’s how far it can push its newfound momentum in an industry where every nanometer matters.