Elon Musk spent the weekend in Austin, Texas, laying out what may be the most expansive vision he has ever attached his name to. The project is called Terafab, a joint venture linking Tesla, SpaceX and his artificial intelligence company xAI, with a single defining goal: to bring every stage of semiconductor production under one roof. That means chip design, fabrication, memory production and packaging, all consolidated in a single facility planned for the north campus of Giga Texas and designed to dwarf even that building, which already ranks among the largest structures on Earth.
The price tag to get started sits between $20 billion and $25 billion, with Musk clarifying that Tesla’s capital expenditure plans for 2026 do not yet account for Terafab costs. The announcement carries added weight given that a SpaceX initial public offering could arrive as early as this spring.
Why Musk says he has no choice
The driving logic behind Terafab is straightforward. Musk has concluded that the world’s leading chipmakers, including industry giants like TSMC and Samsung, are simply not expanding fast enough to meet the demands his companies are generating. Rather than wait for the broader industry to catch up, he intends to build his own supply.
The scale of what he is describing is staggering. Musk stated that current global AI computing output sits at roughly 20 gigawatts per year and suggested that his companies alone need far more than what the rest of the world is currently producing combined. Whether or not those figures hold up to scrutiny, they reflect the urgency with which he is approaching the chip shortage problem.
Two chip types, one enormous demand problem
Terafab is designed around two primary chip categories. The first is an edge-inference processor built to power Tesla’s Full Self-Driving systems, its Optimus humanoid robots and its growing Robotaxi fleet. The second is a hardened high-power variant engineered for space environments, intended to support SpaceX satellites, orbital data centers and xAI operations.
Of these, the Optimus program represents the single most voracious source of demand. Industry analysis projects that Giga Texas alone could eventually hold capacity for up to 10 million humanoid robots per year, a production level that would require roughly 20 million chips, approximately six times what Tesla currently needs across its entire automotive business. If Tesla reaches its stated long-term target of 100 million Optimus units annually, chip demand would exceed 200 million, dwarfing current requirements by a factor of more than 50.
Terafab would pursue a 2-nanometer process, the most advanced fabrication technology in existence and one that established players are only beginning to bring to market at scale.
The case for and against Terafab
Musk’s ambitions have always attracted skepticism, and Terafab is no exception. He has no background in semiconductor manufacturing, a field built on decades of incremental engineering gains that cannot easily be shortcut. Analysts who have looked closely at the project estimate that total capital investment needed to build meaningful capacity could reach between $35 billion and $45 billion, well above the initial figures Musk cited. Even under the most aggressive buildout scenario, first chip output would likely not arrive until mid-2028 at the earliest.
What Musk does have is a track record of doing things the industry said could not be done. Electric vehicles at scale and reusable orbital rockets were both widely dismissed before Tesla and SpaceX made them unremarkable. He leaned directly into that history over the weekend, inviting doubters to revisit their assumptions.
Whether Terafab becomes a genuine turning point in how AI hardware is built, or simply the latest example of a timeline that proves more elastic than expected, the project has already forced a serious conversation about who controls the future of computing.
