Samsung Electronics will start building its second fabrication facility at Taylor, Texas this year and has locked 70% of its memory chip capacity into long-term supply deals, thelec.net reported.

The facility, called Taylor Fab 2, follows Taylor Fab 1, where Samsung broke ground in late 2022. Samsung continues to ramp Taylor Fab 1 toward commercial output, with the two-fab structure splitting the Texas site between different semiconductor processes.

What does the announcement cover?

According to the thelec.net report, Samsung has committed to three actions:

• Beginning construction on Taylor Fab 2 in 2025
• Running memory-focused production lines at the new fab
• Pre-allocating 70% of memory capacity to long-term agreements (LTAs)

Samsung has not disclosed the Fab 2 capex figure or its exact production mix between DRAM and NAND flash. The two product categories split Samsung's memory output, with DRAM serving computing and graphics workloads and NAND serving storage applications.

Why does locking 70% capacity matter?

Memory pricing has historically tracked spot markets, where contract and spot prices swing 40–60% within a single calendar year. Long-term agreements fix pricing tiers and commit volumes over two- to three-year horizons, smoothing revenue through cycle troughs.

Securing LTAs for 70% of memory output ahead of a new fab's ramp provides revenue visibility during the initial yield-ramp phase, when per-wafer costs run high. Similar arrangements have become standard across the memory industry as suppliers hedge against the volatility that defined the 2022–2024 downcycle.

What shifts in the broader memory market?

A 70% pre-booked share indicates that anchor customers hold visibility on demand well beyond typical cycle troughs. For DRAM, AI accelerator supply programs have absorbed much of the HBM3E and HBM4 allocation that suppliers have added since 2023.

For NAND, enterprise storage requirements tied to AI training datasets and inference workloads have consumed most of the additional supply introduced over the past 24 months. Samsung, SK hynix (including Solidigm), Kioxia, and Micron account for nearly all global NAND bit output.

The combined effect pulls incremental capacity out of spot availability, tightening contract pricing power for suppliers and limiting buyers' flexibility to switch vendors mid-cycle.

What questions remain unanswered?

The thelec.net report does not specify which customer segments anchor the 70% LTA book — hyperscale data center operators, smartphone OEMs, SSD vendors, or AI accelerator makers. The report also does not name the contract windows or pricing formula structure.

Taylor Fab 2's total wafer capacity, toolset selection, and timeline for first wafers out also remain undisclosed. Memory fabs typically require 18–24 months from groundbreaking to first silicon, and another 6–12 months to reach commercial yields.

Samsung has not indicated whether the Fab 2 LTA book overlaps with Taylor Fab 1 commitments or stands independently. The Austin, Texas legacy fab, which produces mature logic and foundry products, sits outside the Taylor complex.