Samsung reportedly shifts 2nm production to Texas fab with Tesla as anchor amid yield struggles

Samsung’s foundry unit is expanding its U.S. footprint with the Taylor, Texas fab, slated to trial 2nm production by late 2026. The facility’s anchor tenant, Tesla, signed a $16.5 billion deal for the AI6 chip, but volume production is delayed until late 2027 due to engineering run setbacks. The fab’s scope has narrowed, with U.S. CHIPS Act funding revised downward to $4.745 billion, reflecting both cost pressures and yield uncertainties. For engineers, this means Samsung’s 2nm timeline remains fluid, with potential risks to supply chain commitments if yield targets aren’t met.

Yield remains the critical bottleneck for Samsung’s foundry business. The company’s 2nm yields are reportedly near 55%, below the ~70% threshold typically required for profitability at advanced nodes. This gap has forced Samsung to rely on its memory division’s record profits to absorb foundry losses, a luxury pure-play foundries like TSMC don’t have. The lack of a standalone foundry P&L obscures the unit’s financial health, but the reported 11:1 revenue gap with TSMC underscores the scale of the challenge. Engineers evaluating Samsung for custom silicon must factor in these yield risks, particularly for high-volume designs.

Samsung’s roadmap adjustments reflect broader industry pressures. The company has delayed High-NA EUV deployment for 1.4nm, opting to keep the node on Low-NA tooling, which could slow its ability to close the gap with TSMC. Meanwhile, capex cuts and fab pauses in 2024-2025 signal caution, even as AI memory demand revives expansion plans at Pyeongtaek. The Taylor fab’s 50,000 wafer starts per month target is ambitious, but its success hinges on stabilizing 2nm yields and securing additional customers. For engineers, this means Samsung’s advanced-node roadmap is less predictable than TSMC’s, with potential trade-offs between cost and reliability.

The Tesla deal highlights Samsung’s strategy to lock in anchor tenants for its U.S. fab, but the AI6 chip’s six-month slip underscores the risks of relying on unproven nodes. Samsung’s 2nm family includes SF2 (mass production), SF2P (2025), SF2P+ (Taylor), and SF2Z (2027), but yield instability could delay volume ramp. Engineers targeting AI or automotive silicon must weigh Samsung’s aggressive roadmap against TSMC’s proven track record, particularly for designs requiring high-volume, high-reliability production.