Elon Musk unveils Terafab in bid to unlock massive AI compute in space

Elon Musk has revealed Terafab, a hyper-scale chip production factory aimed at unlocking massive AI compute and ultimately supporting space-based infrastructure and human expansion beyond Earth.

Tesla, SpaceX, and xAI, now part of Musk’s aerospace company, are jointly developing the project. Their goal is to produce one terawatt of computing each year, roughly 50 times the current global output of AI chips.

“The way to actually scale civilization is to scale power in space…because we actually capture such a tiny amount of the sun’s energy on Earth,” Musk said in a recent SpaceX broadcast.

“We want to be a civilization that expands to the galaxy with spaceships that anyone can go anywhere they want at any time,” he added. “To do that, we need to harness the power of the sun. A Terafab, while it is enormous, a terawatt of compute per year is enormous by our sort of civilizational standards, it is still just one step along the way of being even a Kardashev.”

The fabrication facility

Terafab would integrate the full chip development cycle into a single facility, according to Musk. The system would include capabilities for lithography mask creation, chip fabrication, testing, and redesign, enabling a rapid feedback loop to iterate on chip designs.

Musk suggested that this approach could considerably accelerate improvement cycles compared to the fragmented structure of today’s chip supply chain.

The project is expected to begin with an advanced manufacturing facility in Texas, supported by state-level backing.

Two chip categories

The initiative envisions two distinct chip categories. The first would be optimized for edge inference, the kind of onboard processing required by Tesla’s Optimus humanoid robots and its autonomous vehicle fleet, including the forthcoming Cybercab.

Musk projects that humanoid robot manufacturing could eventually reach one billion to ten billion units annually, dwarfing the roughly 100 million vehicles produced worldwide each year.

The second chip variant would be purpose-built for space conditions, engineered to withstand high-energy particle bombardment, and designed to run at elevated temperatures to reduce the mass of thermal radiators on orbiting platforms.

Why space, not Earth

Musk argued that terrestrial power constraints make it physically impossible to deploy a terawatt of compute on Earth, where total US electricity generation hovers around 0.5 terawatts. Instead, the bulk of computing infrastructure would orbit the planet aboard solar-powered AI satellites.

A prototype “mini-satellite” specification calls for 100 kilowatts of output, with future iterations scaling into the megawatt range. Achieving the full terawatt target would require launching roughly ten million tons of material to orbit each year at an efficiency of 100 kilowatts per ton.

The current Starship V3 variant can deliver approximately 100 tons to orbit per payload, a figure that the forthcoming V4 version would double to 200 tons. SpaceX has completed more than 500 successful booster landings and driven launch costs from over $65,000 per kilogram during the Space Shuttle era to an estimated $1,000 to $2,000 per kilogram today.

The company’s stated ambition is to push that figure to between $100 and $200 per kilogram with Starship optimization, a threshold Musk believes would make space-based AI deployment cheaper than ground-based alternatives within two to three years.

To support the transition, Musk pointed to the development of fully reusable launch systems such as Starship, which he said will be critical for transporting the massive volumes of equipment required.

“Starship is a critical piece of the puzzle, because in order to scale compute and scale power, you have to go to space, which means that you need massive payload to space. And Starship will enable that,” Musk said.

He also outlined long-term concepts, including the use of lunar-based manufacturing and mass drivers to further reduce the cost of deploying infrastructure into orbit.

The scale of the gap

Global AI compute capacity is approximately 20 gigawatts per year. Every semiconductor fabrication plant on the planet combined accounts for roughly 2% of what Terafab would need to reach its one-terawatt annual target.

Musk emphasized that existing suppliers, including major global foundries, remain critical, but noted that their maximum comfortable expansion rate falls far short of his requirements.

“We either build the Terafab or we don’t have the chips,” he said. “And we need the chips, so we build the Terafab.”