Elon Musk’s TeraFab: Inside the $25 billion bet to build chips, robots and orbital AI

Elon Musk is attempting one of his most ambitious vertical integration plays yet. With “TeraFab,” a proposed $20-$25 billion semiconductor manufacturing project, the Tesla, SpaceX and xAI chief is looking to bring chip design, fabrication and deployment under one roof, while extending that stack into space.

The goal is straightforward but audacious: reduce dependence on external foundries like TSMC and Samsung, and build a tightly integrated AI hardware ecosystem spanning electric vehicles, humanoid robots, satellites and data infrastructure.

What is TeraFab?

TeraFab is a proposed semiconductor fabrication facility that will design, test and manufacture custom AI chips exclusively for Musk’s companies. The plant is expected to come up on Tesla’s campus in eastern Travis County, Austin, Texas. By making its own chips, Musk wants tighter control over performance, cost and supply, instead of relying on external manufacturers.

Unlike traditional fabless models, where companies design chips but outsource manufacturing, TeraFab aims to internalise the entire stack. That includes advanced packaging, memory integration and inference-optimised silicon tailored for specific use cases across Tesla, SpaceX and xAI.

Chips for cars and robots

The first set of chips from TeraFab will focus on AI inference, chips designed to run AI models quickly and efficiently. These chips will power Tesla’s self-driving cars, robotaxi fleet and the Optimus humanoid robot.  

The chips are expected to be built on a 2nm process, one of the most advanced manufacturing technologies. Smaller nodes usually mean faster performance and better energy efficiency.

The project will also include SpaceX’s D3 chips, which are meant for AI-powered satellites.

The big bet: 1 terawatt of AI in space

What makes TeraFab stand out is Musk’s plan to take computing into space.

He has proposed building large, solar-powered satellites, each about 170 metres long, that can generate around 100 kilowatts of power. These satellites would act like data centres in orbit.

Musk believes solar power is much stronger in space, about five times higher than on Earth. That could make space-based computing cheaper and more efficient over time.

The long-term goal is huge: up to 1 terawatt of AI computing capacity in space. On Earth, the project targets 100–200 gigawatts of compute capacity.

Can Musk out-fab the giants?

Despite the scale of ambition, TeraFab faces many execution challenges. Neither Tesla nor SpaceX has prior experience operating a semiconductor fabrication facility, one of the most complex and capital-intensive industrial processes.

The Musk-led companies will be entering a market dominated by established players like TSMC, which has spent decades refining its manufacturing processes and supply chains.  

There is also the question of ecosystem dominance. Nvidia continues to lead the AI hardware market, with its GPUs forming the backbone of most AI workloads today. Musk’s bet is that vertically integrated, task-specific chips can outperform general-purpose GPUs in targeted applications like autonomous driving and robotics.

Why it matters

If successful, TeraFab could redefine how AI infrastructure is built, shifting from a fragmented, global supply chain to tightly integrated, company-owned stacks that span silicon to software to deployment.

It also signals a broader shift in the AI race, where control over compute, rather than just models, may determine long-term advantage.