Terafab points to a broader Musk push to internalize the AI stack

Terafab is starting to look less like a stand-alone chip project than part of a wider push to bring more of the AI stack inside Elon Musk’s orbit. Based on what has surfaced so far, the effort appears aimed at pulling together chip design, fabrication, compute, data-center capacity and distribution across Tesla, xAI, X and SpaceX.

The near-term outline is concrete enough to show where that push begins. Reuters reported that a pilot fab is planned at Tesla’s Giga Texas site at an estimated cost of about $3 billion, with one facility tied to vehicles and Optimus and another tied to space-based AI data centers. That points to something larger than a custom silicon effort. It suggests a bid to reduce dependency on the outside constraints now shaping AI expansion, including chip supply, power, build timelines and access to advanced compute.

There are also clear signs that outside dependence is still central to the system. Musk said in March that Tesla and SpaceX AI would continue ordering Nvidia chips at scale, even as Tesla moved toward Intel’s 14A process. Intel said Tesla would be its first major 14A customer, and Musk has also named Samsung, Micron, Applied Materials and Tokyo Electron as part of the broader effort. The pattern appears to be less about replacing external suppliers now than about building a path to greater AI infrastructure control over time.

The scale problem

The longer-range vision is where the project becomes harder to map. Reuters has cited estimates of $5 trillion to $13 trillion to reach the stated goal of 1 terawatt of annual compute capacity, with no firm timeline attached. That suggests the financing and execution path remain far from settled. The risk is that the industrial ambition is running ahead of the physical and financial systems needed to support it.

The corporate structure is moving in the same direction. xAI said it acquired X in March 2025, and SpaceX acquired xAI in February 2026, tightening control over data, compute and distribution inside the same network. Terafab appears to fit that pattern, as part of a broader attempt to internalize more of the AI stack across Musk’s companies.

But the end state is still unresolved. Reuters reported, citing SpaceX’s pre-IPO S-1 filing, that the company warned space-based AI data centers rely on unproven technologies and may not become commercially viable. That leaves a gap between the narrative of vertical control and what the underlying system can presently support, with key questions around capital, timelines and technical feasibility still open.