Analysis / Published 2026-11-05

Samsung's second Taylor fab is reportedly booked out, mostly by AMD

Samsung has reportedly booked the entire planned output of its second fab in Taylor, Texas, and most of it is going to AMD. Korea's Maeil Business Newspaper reported the orders, citing industry sources. TrendForce summarised the story; Samsung declined to comment, and Reuters said it could not confirm the report, so the booking is a claim rather than a filing.

(See the reader poll and source context below).

A Korean business paper says AMD has taken most of the output of Samsung's 2nm Taylor Fab 2, four years before the plant is due to start mass production.

Flat editorial illustration of a large circular wafer disc covered in small dark squares at the left, and a low grey factory building with two linked chip squares above it at the right.
Conceptual editorial illustration of a wafer fab booked ahead of construction; it is not a company photo and shows no real facility.

What do the sources establish?

Recorded source context. No verified social reaction quotes or aggregate vote results are supplied.
Platform / SourceCore Stance / QuoteEngagement MetricTimestamp
Samsung's 2nm Taylor Fab 2 reportedly fully booked before 2030 output, with AMD leading HBM base die ordersReported context: [object Object] [object Object] [object Object] [object Object]Not recorded
Samsung's next Texas fab could be almost entirely for AMDReported context: [object Object] [object Object] [object Object] [object Object]Not recorded

THE SHORT VERSION

  • Maeil Business Newspaper reported that Samsung has secured orders covering the entire planned capacity of Taylor Fab 2, with AMD taking the majority, though Samsung declined to comment and Reuters could not confirm it.
  • The plant would make HBM base dies on a 2nm process in Taylor, Texas, with construction starting this month, equipment arriving from February 2029 and mass production in 2030.
  • A fab booked before ground is broken would mark a turnaround for Samsung's loss-making foundry unit, but Nvidia's reported preference for TSMC base dies limits how much of the upside Samsung controls.

The report, and its caveats

Samsung has reportedly booked the entire planned output of its second fab in Taylor, Texas, and most of it is going to AMD. Korea's Maeil Business Newspaper reported the orders, citing industry sources. TrendForce summarised the story; Samsung declined to comment, and Reuters said it could not confirm the report, so the booking is a claim rather than a filing.

The sourcing matters here. A fab that is filled before ground is broken is exactly the kind of claim a foundry wants repeated, and Samsung's foundry unit has spent years losing money. The details that can be checked, the site, the process and the schedule, are consistent across the Korean and English reports, which is not the same as confirmation, and readers should treat the order book as unverified.

Base dies, in plain terms

The plant will produce base dies for high-bandwidth memory, the logic layer that sits at the bottom of the stacked memory chips feeding AI processors. Today's HBM4 and sampled HBM4E parts use 4nm base dies; Taylor Fab 2 would make them on 2nm or better for the next generation of memory. The die carries the memory controller and the interface that a GPU sees.

A base die is not the memory itself. It is the control logic that talks to a GPU, which is why it is built on a logic process rather than a memory one, and why a memory maker needs a foundry to make it. Samsung is one of the few companies that does both under one roof, which gives it a second lever when memory supply is tight.

The Taylor timetable

Preparatory work, including site preparation and ground levelling, is already underway, with full-scale construction scheduled to begin in early November 2026, according to the report. Equipment installation is expected to start in February 2029, followed by mass production in 2030, which is a long way from a booking today. The gap between order and output is the story.

The building will span 2.7 million square feet, about 250,000 square metres and roughly 1.5 times the size of the first Taylor fab. Samsung plans to pair base dies from Texas with DRAM produced in South Korea to ship completed HBM stacks, which keeps the memory and the logic in separate factories on separate continents, and lets each be qualified and expanded on its own schedule.

Why AMD wants the capacity

AMD has been locking in capacity across its supply chain as its AI chip business scales, and memory has been the tightest part of that chain. Base dies on 2nm would give AMD a generation of HBM that does not depend on TSMC's logic capacity, which is the constraint every accelerator vendor is bidding against, and whoever locks it early avoids a surprise at delivery.

The volumes also fit AMD's position. It does not need the largest allocation in the industry, only enough to keep accelerators shipping, and AMD's data center quarter showed how much of its growth now rests on that. A fab booked mostly by one customer is a supply agreement dressed as a property story, where predictability matters more than scale.

Samsung's foundry turnaround

For Samsung, a second fab filled before construction would mark a turnaround for a foundry arm that has lost money for years. Kiwoom Securities expects the unit to return to profit from the fourth quarter, and Samsung has projected that its 2nm orders will grow more than 130% this year, helped by AI demand and tight capacity at TSMC. If that forecast holds, the loss curve bends this quarter.

The first Taylor plant, which grew from $17 billion to a $37 billion total investment, is due to start mass production in 2027, with Tesla as its anchor customer under a $16.5 billion contract for AI5 and AI6 chips. The site covers 1,268 acres, has room for up to 10 fabs and took a $4.75 billion subsidy under the US CHIPS Act, which is why a second building was always going to follow the first.

The Nvidia question

The reported orders do not settle Samsung's place in Nvidia's next-generation HBM supply chain. Maeil reports that Nvidia has asked all three memory makers to outsource NVHBM base die production to TSMC, which would leave Samsung building memory around someone else's logic, and that limits what a fully booked fab is worth, because the value sits in the die as much as in the stack.

There are other names in the frame. Industry observers see Broadcom as a possible customer for base dies from Taylor Fab 2, and ChosunBiz reported that Samsung could make 2nm AI chips for a US cloud customer through Faraday, an ASIC designer in Taiwan. Neither order has been confirmed by anyone involved, and neither has a product attached that a customer could point to.

What to watch next

The first checkpoint is the start of construction this month, which is visible and hard to walk back. The second is whether AMD or Samsung puts a number on the booking, because a capacity agreement that cannot be described in a filing is still a report, however specific it sounds, and a filing would show how much of the queue is AMD and who else is in it.

The third is TSMC's capacity. If it stays short, customers will pay for a second source, and Samsung's 2030 output arrives into that gap; if it loosens, the case for a fab booked by one customer weakens. This site has tracked the scale of the buildout and the infrastructure earnings that this memory is meant to serve.

Sources & reporting limits

  1. Samsung's 2nm Taylor Fab 2 reportedly fully booked before 2030 output, with AMD leading HBM base die orders
  2. Samsung's next Texas fab could be almost entirely for AMD
Keep the thread.

Follow Industry to find related stories on your next visit.

THE WEEKLY EDIT

The week, made clear.

Stories worth understanding, brought together in one weekly edition.

Share this page

Instagram, TikTok & other apps

Copy the page link, then paste it where your chosen app supports links. You choose whether to send or publish.