Elon Musk Prepares His Own Chip Fab: Tesla and SpaceX Launch the TeraFab Project

Elon Musk wants Tesla and SpaceX to stop depending on outside suppliers and make their own critical chips. This is not just about cars, but also about robots, AI infrastructure and even orbital data centers.

A fab of their own

The project brings together two of Musk's key companies: Tesla and SpaceX. They are supposed to jointly launch an advanced chip-manufacturing site in Austin because, in Musk's view, the current semiconductor market is expanding capacity too slowly for his needs. He needs processors not for a single product line but for several areas at once: robotics, model training and inference, and the computing systems that SpaceX ties to its space plans.

The defining feature of the idea is an attempt to gather almost the entire chain under one roof. Discussion of the project covered logic, GPUs and CPUs, memory and packaging. For the industry this is an unusual move: these stages are normally split among different players because the technologies, materials and process economics differ greatly. So what is unusual here is not only the size of the project but also the bet on extreme vertical integration.

"Either we build TeraFab, or we won't have chips.

We need chips, so we build TeraFab."

A bet on scale

The scale of the project is almost demonstrative. The starting target is roughly 100,000 wafers per month, and the long-term goal is to reach 1 million per month. That no longer looks like an in-house fab for the narrow needs of one company. It looks more like an attempt to build a proprietary industrial base for AI, robotics and space. At this scale, the main question is not a polished presentation but money, timelines and the ability to assemble the necessary engineering expertise quickly.

• The starting target is about 100,000 wafers per month
• The long-term target is 1 million wafers per month
• The plan is to combine logic, memory and packaging on a single site
• One of the main use cases is chips for space-based data centers

A separate layer of the story is the link to orbital computing. The project gives a significant role to space data centers, which SpaceX has long pitched as the next big market for AI compute. The public presentation also showed concepts of satellite computing platforms. If Starship really does get cheaper and closer to full reusability, the economics of launching such systems could change noticeably. For now it looks very ambitious, but the logic is clear: control not just the models and servers but the whole stack, from chip to orbit.

The infrastructure race

Another front is accelerating in parallel: financing for AI infrastructure. In March, UK-based Nscale raised $2 billion in a Series C round and reached a $14.6 billion valuation. Sheryl Sandberg joined its board of directors, along with Susan Decker and Nick Clegg. For the European market this is an important signal: big money is flowing not only into models and applications, but also into companies that build the computing base, data centers, GPU clusters and services on top of them.

Where Musk's strategy is built around in-house manufacturing, Nscale is betting on the speed of deploying infrastructure across regions and on the weight of a heavyweight board. Both cases point to the same fork in the market: in AI there is less and less room for lightweight intermediaries, and more and more value for those who control scarce resources: energy, sites, accelerator supply and production chains.

What this means

The AI market is shifting quickly from a race between models to a race for physical infrastructure. If Musk's plan works, Tesla and SpaceX will try to turn the chip shortage into a competitive advantage, while Nscale's success shows that investors are ready to pay huge sums for control over computing capacity.