Terafab: Elon Musk's massive factory will build AI chips for cars, robots, space projects
Elon Musk on Saturday unveiled Terafab, terming it a groundbreaking semiconductor project jointly led by Tesla and SpaceX near Austin, Texas.
What is Terafab?
This "advanced technology fab" aims to produce one terawatt of computing power annually, focussing on custom AI chips for Tesla's vehicles, robotaxis, Optimus robots, SpaceX satellites, and xAI models.
A terawatt is equivalent to one trillion watts. That is slightly less than the total power generation capacity of the United States, according to an industry group.
It will manufacture two chip types: edge/inference-optimised for Tesla's Full Self-Driving, Cybercab, and Optimus; and high-performance chips for xAI training and SpaceX orbital data centres.
Initial costs of the project are estimated at $20-25 billion. The facility integrates design, production, testing, and scaling, potentially using Optimus robots for operations. Plans include space-based AI via millions of satellites, leveraging vacuum cooling and constant solar power for cheaper orbital compute.
Musk's plan for Terafab
The project will start small in Austin, and scale to 100-200 gigawatts terrestrial and one terawatt orbital power, enabling "galactic civilisation" by powering AI from Earth to space. xAI, now a SpaceX subsidiary, will get priority output.
Musk did not give a timeline for Terafab's output, and has previously promised grand results from other projects on compressed time scales, but did not always deliver.
Why now?
"We need the chips, so we build the Terafab," said Musk. The plant is his response to surging AI chip demand outpacing suppliers like TSMC and Samsung.
Musk has no prior experience in semiconductors. But he said the Terafab was necessary because Tesla and SpaceX's demand for computing power was expected to far exceed that of global chip suppliers.
The latest announcement also comes shortly after SpaceX indicated a shift in near-term priorities from Mars to the Moon, suggesting that its plans for a Mars colony may take more time to materialise than previously anticipated.
Broader ambition
Elon Musk's push into custom chips started with Tesla's Dojo supercomputer, launched in 2021 to train Full Self-Driving (FSD) software using proprietary D1 chips. In January 2025, Tesla announced a pivot from separate chip designs, unifying on AI5 chips, launched last July with performance rivalling Nvidia's Hopper GPUs but using far less power. AI6 chips followed in April 2026, under a rapid 9-month release cycle to keep pace with autonomy needs.
This in-house strategy will slash the EV major's dependence on Nvidia and others.
What is Terafab?
This "advanced technology fab" aims to produce one terawatt of computing power annually, focussing on custom AI chips for Tesla's vehicles, robotaxis, Optimus robots, SpaceX satellites, and xAI models.
A terawatt is equivalent to one trillion watts. That is slightly less than the total power generation capacity of the United States, according to an industry group.
It will manufacture two chip types: edge/inference-optimised for Tesla's Full Self-Driving, Cybercab, and Optimus; and high-performance chips for xAI training and SpaceX orbital data centres.
Initial costs of the project are estimated at $20-25 billion. The facility integrates design, production, testing, and scaling, potentially using Optimus robots for operations. Plans include space-based AI via millions of satellites, leveraging vacuum cooling and constant solar power for cheaper orbital compute.
Musk's plan for Terafab
The project will start small in Austin, and scale to 100-200 gigawatts terrestrial and one terawatt orbital power, enabling "galactic civilisation" by powering AI from Earth to space. xAI, now a SpaceX subsidiary, will get priority output.
Musk did not give a timeline for Terafab's output, and has previously promised grand results from other projects on compressed time scales, but did not always deliver.
Why now?
"We need the chips, so we build the Terafab," said Musk. The plant is his response to surging AI chip demand outpacing suppliers like TSMC and Samsung.
Musk has no prior experience in semiconductors. But he said the Terafab was necessary because Tesla and SpaceX's demand for computing power was expected to far exceed that of global chip suppliers.
The latest announcement also comes shortly after SpaceX indicated a shift in near-term priorities from Mars to the Moon, suggesting that its plans for a Mars colony may take more time to materialise than previously anticipated.
Broader ambition
Elon Musk's push into custom chips started with Tesla's Dojo supercomputer, launched in 2021 to train Full Self-Driving (FSD) software using proprietary D1 chips. In January 2025, Tesla announced a pivot from separate chip designs, unifying on AI5 chips, launched last July with performance rivalling Nvidia's Hopper GPUs but using far less power. AI6 chips followed in April 2026, under a rapid 9-month release cycle to keep pace with autonomy needs.
This in-house strategy will slash the EV major's dependence on Nvidia and others.
Next Story