Samsung Taps AMD and Nvidia for HBM Base Dies at Second Texas Fab

Samsung is doubling down on Texas. A second foundry at its Taylor site — dubbed Taylor Fab 2 — will churn out high-bandwidth memory base dies exclusively for AMD and Nvidia, according to research findings on the plan. Groundbreaking is set for early November 2026, with production expected to start in 2030.

The scale is serious. Fab 2 will span roughly 2.7 million square feet, making it about 1.5 times the size of Taylor Fab 1, which is still under construction. The base die is the logic layer at the bottom of an HBM stack, handling the traffic between the memory above and the AI processor doing the heavy lifting.

Samsung is splitting the work across two countries. Base dies get made in Texas, then get paired with DRAM core dies produced in South Korea to form complete HBM stacks. Current HBM4 tech runs base dies on a 4nm process; Fab 2 is aiming for 2nm or beyond.

The two-plant strategy is coming into focus. Fab 1 is anchored by a Tesla supply deal for next-generation AI chips valued at roughly $16.5 billion, with full operational capacity expected by the end of 2026 and a ramp toward production in 2027. Fab 2 is anchored by AMD and Nvidia — Samsung is effectively pre-selling its Texas capacity to American tech's biggest buyers before the concrete is poured.

The competitive read is obvious. TSMC dominates the foundry market, and locking in orders from AMD and Nvidia gives Samsung a more credible claim as an alternative. The research also frames the expansion as a boost to US capacity for a critical AI component, potentially easing reliance on Asian producers.

The hybrid model muddies that story a bit — DRAM core dies still ship in from South Korea, so the finished stacks are a transpacific product. The near-term test sits at Fab 1: whether it hits full capacity by the end of 2026 and ramps on schedule in 2027 will tell the market plenty about Samsung's ability to execute in Texas.