- SpaceX filed for a $55 billion semiconductor fab in Texas that could reach $119 billion total, aiming to produce 1 terawatt of computing power annually for AI chips.
- The Terafab project would combine with existing Bastrop packaging operations to create the largest PCB and panel-level-packaging facility in North America.
- Morgan Stanley estimated $45 billion; the actual projection is 2.6x higher, with County Commissioners meeting June 3 on tax abatements.
SpaceX filed paperwork this week for a semiconductor fabrication facility in rural Texas with an initial investment of $55 billion that could swell to $119 billion across multiple phases, according to a proposal on the Grimes County website, covered by TechCrunch. The project, dubbed Terafab, would sit alongside the company’s existing chip-packaging operations in Bastrop and anchor a Texas chipmaking footprint that Musk claims will eventually produce enough silicon to deliver 1 terawatt of computing power per year.
The filing comes as SpaceX prepares for what’s expected to be the largest IPO in history, with the combined SpaceX-xAI entity valued at $1.25 trillion. Musk has previously said the facility will supply chips for AI servers, satellites, SpaceX’s proposed data center in space, and autonomous Tesla vehicles and robots. The companies have roped Intel into the effort, though the specific process technology Terafab will run remains undisclosed.
Morgan Stanley analysts estimated in March that a factory capable of producing chips at the scale Musk had suggested could cost up to $45 billion. The $55 billion initial figure and $119 billion total projection in the Grimes County filing significantly exceed those estimates. County Commissioners will meet on June 3 to consider approval of a property tax abatement for the project.
The Economics of Owning the Stack
The distinction between Terafab and the existing Bastrop facility matters. Bastrop is a packaging operation—it takes silicon dies fabricated elsewhere, packages them, and ships them out as finished radio-frequency chips for Starlink user terminals. Terafab is something different: an actual fabrication facility designed to produce silicon at process nodes rather than only package it.
Packaging is a smaller, faster, more capital-light business than fabrication. A fab requires cleanroom facilities, lithography equipment that costs hundreds of millions per unit, deep semiconductor process expertise, and construction cycles that typically run five to seven years. The $55 billion Terafab figure is consistent with that scale of project, and the combined Bastrop and Terafab footprint would become the largest PCB and panel-level-packaging facility in North America.
The strategic rationale fits SpaceX’s broader pattern. Starlink ships hardware at volumes that make per-unit silicon cost a non-trivial line in the company’s economics. Owning the fab, the packaging, and the PCB manufacturing in a single integrated US facility removes the third-party supplier margin from each step in the production chain and gives SpaceX direct control of the supply timeline for components that cannot be substituted late in the build cycle.
What 1 Terawatt Actually Means
Musk’s claim that Terafab will eventually manufacture enough chips to provide 1 terawatt of power per year is worth unpacking. A terawatt is 1 trillion watts, or roughly 100 times the total generating capacity of the United States. But in the context of semiconductor manufacturing, the metric refers to the aggregate compute power of the chips produced, not electricity generation.
Nvidia’s H100 GPU draws roughly 700 watts at full load. If Terafab were producing H100-class chips exclusively, 1 terawatt of annual output would imply roughly 1.4 million chips per year. That’s roughly double Nvidia’s entire H100 production capacity in 2025. But Musk’s companies need more than GPUs—they need custom ASICs for Starlink terminals, specialized processors for autonomous driving, and AI accelerators optimized for xAI’s Grok models.
The 1 terawatt figure is aspirational rather than a near-term production target. Semiconductor fabs typically ramp up over years, not months. TSMC’s Arizona fab, which broke ground in 2021, won’t reach full 5nm capacity until 2028. Terafab’s timeline is similarly opaque—SpaceX has not publicly committed to a construction start or operational date, only to the filing.
The Geopolitics of Domestic Manufacturing
Texas Governor Greg Abbott’s office has been actively supporting the Bastrop expansion. A March 2026 announcement from the governor’s office confirmed a Texas Semiconductor Innovation Fund grant to SpaceX. Terafab would be eligible for additional state-level incentives at scale, alongside any federal CHIPS Act support that survives the current administration’s review.
The geographic logic fits a wider US pattern. Apple’s recent foundry-diversification discussions with Intel and Samsung are the demand-side counterpart, and Intel’s hire of a senior Qualcomm executive to lead its new Client Computing and Physical AI Group is the design-side complement. Texas in particular has been attracting AI-infrastructure commitments at scale—Hut 8’s $9.8 billion Beacon Point lease in Nueces County is the parallel data-centre commitment in the same state.
The proposed location for the Terafab site is the Gibbons Creek Reservoir in rural Texas, about 100 miles from Tesla’s headquarters in Austin and close to College Station, home of Texas A&M University. The reservoir location is strategic—semiconductor fabs require massive amounts of water for cooling, and proximity to a university system helps with the specialized technical talent that analysts say the project will struggle to recruit.
The Open Questions
Two questions remain unresolved on the public record. The first is which specific process technology Terafab will run. SpaceX has not disclosed it, and the most ambitious leading-edge nodes would require licensing or partnership with an existing process-IP holder. The most likely initial focus is on advanced packaging and mature-node fabrication for Starlink-specific RF and ASIC components rather than on competing directly with TSMC at the leading edge.
The second is timeline. SpaceX has not publicly committed to a Terafab construction start or operational date, only to the filing. Semiconductor fabs are notorious for delays and cost overruns. Intel’s Ohio fabs, announced in 2022 with a $20 billion initial investment, have seen their projected completion dates slip by years and their cost estimates balloon. Terafab’s $119 billion projection assumes everything goes right—a rare outcome in semiconductor manufacturing.
What is no longer in doubt is intent. SpaceX has, on Wednesday’s filing, committed to manufacturing its own silicon at Tier-1 fab scale. Whether the timeline holds, whether the process technology question is answered through partnership or in-house development, and whether the combined $119 billion capital programme survives the next several quarters of competing demands on SpaceX’s balance sheet are the open questions. The filing itself is the news.
FAQ
What is Terafab?
Terafab is SpaceX’s proposed semiconductor fabrication facility in Grimes County, Texas. It’s designed to be a “multi-phase, next-generation, vertically integrated semiconductor manufacturing and advanced computing fabrication facility” that would produce chips for SpaceX, xAI, and Tesla.
How much will Terafab cost?
The initial investment is projected at $55 billion, with total investment potentially reaching $119 billion across multiple phases. This significantly exceeds the $45 billion estimate from Morgan Stanley analysts.
What chips will Terafab produce?
Terafab aims to supply AI chips for Musk’s companies, including processors for Tesla’s robotaxis and Optimus robot, Starlink user terminals, SpaceX’s proposed space-based data center, and xAI’s Grok AI models.
When will Terafab be operational?
SpaceX has not publicly committed to a construction start or operational date. Semiconductor fabs typically require five to seven years from groundbreaking to full production.
What is the 1 terawatt claim?
Musk has said Terafab will eventually manufacture enough chips to provide 1 terawatt of computing power per year. This refers to the aggregate compute power of the chips produced, not electricity generation. It’s an aspirational target rather than a near-term production goal.
