Tesla and SpaceX CEO Elon Musk claims he will build a $20 billion ‘TeraFab’ semiconductor plant to make chips for space data centers, advanced vehicles, and humanoid robots.

Musk announced the plan on Saturday, revealing that the TeraFab would be built in Austin, Texas, adjacent to Tesla’s headquarters, and be run as a joint venture between the automaker and SpaceX.

Describing the project “the most epic chip-building exercise in history by far,” he said his companies require their own manufacturing facilities because current semiconductor production levels are not sufficient to meet their needs. “We either build the TeraFab, or we don't have the chips, and we need the chips, so we build the TeraFab," the billionaire said in a live stream broadcast on his social media platform, X.

He said his firms would eventually need up to 200GW per year of compute power for terrestrial operations, and up to 1TW for space-based processing. Musk claimed in the broadcast that there is currently 20GW of AI compute power available globally, though other estimates put this figure at 30GW.

According to Musk, the TeraFab will make two different types of chip - one for Edge inference workloads, suitable for vehicles and robots, and another, more powerful, line of semiconductors, hardened for space environments.

The space chips will be needed for SpaceX’s planned network of orbital data centers. In January, the firm filed for permission to launch a constellation of up to one million data center satellites, and Musk said at the time the project could provide hundreds of gigawatts of compute.

The TeraFab is expected to cost $20bn, and Musk said it will be a fully integrated facility, capable of making logic and memory chips, as well as undertaking packaging and testing.

Musk has no experience in the notoriously complex and capital-intensive semiconductor manufacturing industry, and a long history of making big promises about his companies that don’t come to fruition. No timeline was put on when the TeraFab would be operational.

He first floated the idea of building his own chip fab in November, and said in January that his team would be aiming for a nine-month production cycle for its semiconductors, three months less than the yearly cadence Nvidia and AMD work to when producing their AI accelerators.

Tesla has previously designed its own chip, the Dojo D1, which was built to process videos from its cars and train its autonomous fleet, and had been slated for use in the company's Optimus humanoid robot. The Dojo team was dissolved in August 2025, with Tesla instead focusing on deploying existing chips from other vendors.

DCD reported exclusively in July 2025 that another of Musk’s companies, AI lab xAI, was hiring staff to develop custom silicon.