Samsung Splits Texas Fabs: Space Chips in Austin, AI Chips in Taylor

By Marcus Bennett

Samsung will make Starlink satellite chips at its Austin fab and advanced AI chips at its Taylor plant, backed by $4.745 billion in U.S. CHIPS Act subsidies. The U.S. Commerce Department allocated the funds, about 6.36 trillion won, toward Samsung's $37 billion (49.6 trillion won) investment in the two Texas sites.

The move turns Samsung's Texas operations into a foundry hub serving both the space industry and artificial intelligence. The strategy aims to expand the scale and customer base of Samsung's contract chip manufacturing inside the home market of U.S. Big Tech.

What technology is involved?

Reporting by The Seoul Economic Daily confirmed that the Commerce Department's key Chips for America projects include fully depleted silicon-on-insulator (FD-SOI). This is the production technology for the satellite chips Samsung will supply to SpaceX through STMicroelectronics.

FD-SOI is a manufacturing process Samsung and STMicroelectronics have jointly developed since 2014. In conventional chips, current generated in transistors can leak toward the silicon substrate and waste power. FD-SOI instead places an extremely thin insulating layer between the silicon layer and the substrate, then forms transistors on top of it. The design suppresses leakage current.

Because it cuts power consumption while improving circuit performance, the process suits aerospace, automotive and industrial equipment. These are fields where power supply and heat management are tightly constrained.

The concrete product is the STM32V8, a satellite microcontroller unit built on FD-SOI. Its central processing unit is an Arm Cortex-M85 running at up to 800 megahertz — 800 million cycles per second. The chip delivers high computing performance and durability at lower power draw.

SpaceX has endorsed the part. Michael Nicolls, vice president of Starlink engineering at SpaceX, said the STM32V8's computing performance and embedded memory were important in meeting demanding real-time processing requirements.

Where will production happen?

The main production base for the STM32V8 is currently STMicroelectronics' plant in Crolles, France. The chip is made on 300-millimeter wafers, with Samsung's foundry taking part in manufacturing.

Under its agreement with the Commerce Department, Samsung will secure capacity for an 18-nanometer FD-SOI process at the Austin plant. That extends Starlink satellite chip production to Austin alongside Crolles.

The Commerce Department said it will support expansion of existing facilities to enable production with leading-edge FD-SOI process technology needed by key U.S. industries, including aerospace, defense and automotive. The department expects all facilities to be operational by 2030.

Once satellite chip output starts at Austin around 2030, Samsung's Texas plants will serve as a strategic base for core chips supplied to Tesla and SpaceX. The structure is a clean division of labor: Taylor handles advanced processes, Austin handles specialty processes. Together they will supply customized chips to U.S. Big Tech firms including Tesla, SpaceX, Nvidia and AMD.

What does this change for Taylor?

The Taylor plant is accelerating its own expansion. Samsung is investing to produce Tesla's AI5 chip for self-driving and robotics on a 2-nanometer process. The next-generation AI6 chip is also slated for Taylor. Tesla has set production targets of 2027 for AI5 and 2028 for AI6.

Industry officials expect the completed expansions to improve Samsung's chances of winning additional orders from U.S. Big Tech firms.

"When local production facilities expand, everything from design changes to prototype validation and technical discussions can be handled nearby, which speeds up development and reduces logistics burdens," an industry official said.

"With Samsung's advanced foundry capacity in Korea getting tight, using the U.S. plants gives it more room to absorb new orders," the official added.

The practical effect is simple. Samsung gains U.S.-based capacity for both cutting-edge AI silicon and low-power space chips, while American customers gain a nearby supplier for design iterations and prototyping.