×
November 16, 2026
Muskette
4 min Read

Samsung Commits $1.9B to Austin Fab for SpaceX Starlink Chips

Listen on the go

Voicing this story…

Samsung is putting serious money into making the tiny brain that lets Starlink satellites chat with each other across the sky. South Korean business daily Seoul Economic Daily reported on November 15 that the company will spend about $1.9 billion expanding its long-running Austin, Texas fab specifically so it can turn out the STM32V8 microcontroller for SpaceX.

That chip sits at the heart of the laser inter-satellite links. Instead of bouncing every packet down to a ground station, the constellation passes data sideways from bird to bird. Real-time processing under tight power budgets and constant radiation is brutal. The STM32V8 handles it with an Arm Cortex-M85 core clocked up to 800 MHz plus embedded memory, all built on an 18-nanometer fully depleted silicon-on-insulator process that Samsung and STMicroelectronics co-developed.

Right now the parts come out of ST’s Crolles plant in France, with Samsung Foundry already involved. The Austin move does not shut that line down. It adds a second, U.S.-based source. Production at the expanded Texas site is targeted for 2030.

TWO TEXAS TRACKS

Samsung is running a clean split in the Lone Star State. The newer Taylor campus is the advanced-node house. That is where Tesla’s AI5 and upcoming AI6 inference chips are headed, along with work for Nvidia and AMD. Austin stays on specialty and mature processes—from 65 nm down to 14 nm—and now gets the FD-SOI capacity the Commerce Department specifically called out for aerospace, defense, and automotive customers.

The bigger envelope is enormous: more than $37 billion across the Texas projects, backed by up to $4.745 billion in CHIPS and Science Act money. NIST materials list the FD-SOI expansion as one of the supported pieces. Samsung has not publicly confirmed the Starlink contract by name—standard foundry practice—but the reporting lines up with earlier ST disclosures that SpaceX selected the STM32V8 for the mini laser system and with comments from Starlink engineering vice president Michael Nicolls praising its compute and memory for the job.

Local production matters for more than flags and subsidies. Design tweaks, prototype spins, and technical meetings become same-day affairs instead of transatlantic logistics. Korean advanced capacity is already tight; the U.S. plants give Samsung room to take more volume without choking the home fabs.

WHY THE TIMING LANDS NOW

Starlink is no longer a few thousand satellites. The network sits above 11,000 birds in orbit with V3 units rolling out on Starship, each packing far higher capacity and bigger solar arrays. Laser mesh traffic only grows as the constellation densifies and as direct-to-cell and enterprise loads climb. A dual-continent supply of the radiation-tolerant MCU that keeps the optical crosslinks honest is the kind of boring infrastructure move that suddenly looks strategic.

It also threads neatly through the Musk ecosystem. Tesla already has a multi-billion-dollar AI chip deal with the same Samsung Texas complex. SpaceX gets its space-grade silicon closer to home. The same state that hosts Starbase, Giga Texas, and the emerging AI campuses now hosts more of the silicon that keeps the satellites talking.

None of this means Starlink terminals get cheaper next month or that download speeds jump overnight. It is a multi-year capacity build. But the fact that the report landed this weekend and spread quickly across tech and space desks shows how closely people are watching every supply-chain decision that touches the constellation. When the lasers need more brains, Samsung just said it will grow them in Austin.

The expansion still needs to clear the usual construction and qualification hurdles before the first Texas-made STM32V8 flies. For now the plan is clear: France keeps running, Texas ramps specialty space silicon by the end of the decade, and Starlink’s optical backbone gets a second source right in the middle of the American semiconductor corridor.

0 COMMENTS

LEAVE A GRITTY RESPONSE