Samsung's Second Texas Fab to Produce HBM Base Chips for AMD and Nvidia
nashnova research
Samsung Electronics will build a second wafer fab in Taylor, Texas, using its 2nm process to produce HBM base dies for AMD and Nvidia — all capacity is already spoken for, with AMD taking the lion's share and mass production targeted for 2030.
What exactly will this new fab make?
The Taylor second fab will produce base dies for HBM (high-bandwidth memory) chips — the bottom layer of an HBM stack that handles communication with the processor.
The confirmed customers are AMD and Nvidia. This means → two of the biggest AI-chip makers are entrusting a critical HBM supply-chain link to Samsung foundry.
All capacity has been locked in by orders, most of it from AMD. In plain terms = the fab is sold out before construction is even complete.
How big is the fab, and when does it come online?
Construction begins this month. The fab uses Samsung's 2nm advanced process and covers roughly 250,000 square meters.
That is about 1.5× the size of Taylor's first fab, which has a contract with Tesla — making it one of Samsung's largest new lines in the US.
Equipment installation is set for February 2029; mass production is targeted for 2030. This means → the build-to-ship cycle runs roughly five years.
Why is AMD betting so much capacity on Samsung?
AMD has secured the majority of the fab's output. This reflects AMD's push to lock down HBM supply-chain resources ahead of its next-generation AI accelerators.
SK Hynix currently dominates HBM manufacturing. In plain terms = AMD does not want a single-supplier dependency and is opening a second supply lane through Samsung.
If the timeline holds, AMD's reliance on Samsung foundry will deepen significantly, binding the two companies closer together.
What does this mean for Samsung itself?
The second Taylor fab marks a major expansion milestone for Samsung's US foundry business and the first large-scale overseas deployment of its 2nm process.
Taylor fab one serves Tesla; fab two is locked in by AMD and Nvidia. This means → Samsung is using a "one city, two fabs" layout to cover both automotive and AI chips simultaneously.
Samsung Electronics declined to comment; the report cites industry sources quoted by South Korea's Maeil Business Newspaper.
市场有风险，内容仅供研究参考，不构成投资建议。