Key Takeaways
- Elon Musk announced plans for Terafab to generate one terawatt of annual AI computing capacity.
- The initiative represents a collaborative effort between Tesla, SpaceX, and xAI.
- The Texas facility would integrate design, manufacturing, testing, and redesign operations.
- Musk envisions the majority of future AI computing operating on solar-powered orbital satellites.
- The Starship launch system plays a crucial role in transporting AI infrastructure to space.
Elon Musk has revealed details about Terafab, an ambitious semiconductor manufacturing facility designed to dramatically scale artificial intelligence computing capacity. The initiative connects chip production with orbital infrastructure and future space-based computing platforms.
The project involves collaboration across Tesla, SpaceX, and xAI. Musk outlined plans to achieve one terawatt of annual computing output, representing a significant leap beyond current AI chip manufacturing levels.
Integrated chip manufacturing under one roof
Musk explained that Terafab would consolidate the entire semiconductor development cycle within a single location. This encompasses mask generation, chip fabrication, performance testing, and iterative design improvements. The approach aims to accelerate development cycles and enhance chip optimization speed.
Texas has been selected as the location for the initial production facility. State-level support is anticipated for the project. Musk criticized existing semiconductor supply chains as overly fragmented and inadequate for achieving the desired production volumes.
Elon Musk said his Terafab project — a grand plan to eventually manufacture his own chips for robotics, artificial intelligence and space data centers — will be built in Austin and jointly run by Tesla and SpaceX https://t.co/c0bXI7mLZt
— Bloomberg (@business) March 22, 2026
Emphasizing the project’s importance, he stated, “We either build the Terafab or we don’t have the chips.” This comment underscores both the project’s magnitude and anticipated future demand levels.
The manufacturing strategy focuses on two distinct chip categories. One type would power edge inference applications for Tesla vehicles, Cybercab transportation, and Optimus humanoid robots. The second category targets space deployment, engineered to withstand extreme thermal fluctuations and radiation exposure.
Orbital computing infrastructure as the foundation
Musk contends that terrestrial power infrastructure cannot support one terawatt of computing capacity. According to available information, total United States electricity generation reaches approximately 0.5 terawatts, falling short of the project’s requirements.
Given these constraints, Musk indicated that the bulk of computing operations would need to function in orbital environments. He outlined solar-powered AI satellites as the primary infrastructure for this expansion. Initial prototype mini-satellites would generate 100 kilowatts of power.
Subsequent iterations would expand capacity into megawatt territory. Achieving the complete objective would necessitate launching approximately ten million tons of hardware into orbit annually. This calculation assumes 100 kilowatts of computing capability per ton of payload.
Highlighting the connection, Musk stated, “Starship is a critical piece of the puzzle.” He positioned the launch vehicle as essential to realizing expanded computing and power capabilities in space.
Launch vehicle capabilities and economics
The existing Starship V3 configuration can transport roughly 100 tons per mission to orbital altitude. Musk projected that the upcoming V4 variant would double this capacity to 200 tons per launch.
According to available data, SpaceX has successfully executed over 500 booster recovery operations. Launch economics have dramatically improved from exceeding $65,000 per kilogram during the Space Shuttle program to approximately $1,000 to $2,000 per kilogram currently.
Musk outlined SpaceX’s objective to further decrease costs to $100 to $200 per kilogram. He projects this price point could enable space-based AI systems to achieve cost parity with terrestrial alternatives within two to three years.
Current worldwide AI computing capacity stands at roughly 20 gigawatts annually. While Musk acknowledged the continued relevance of established chip suppliers, he noted their expansion trajectory falls short of his targets. Terafab represents his proposed solution to address this capacity shortfall.
