Terafab: Tesla, SpaceX, and xAI's chip mega-factory is targeting 200 billion AI chips a year
Elon Musk's companies are building their own semiconductor factory — and the stated ambition is to produce more AI chips annually than the entire current global AI computing fleet combined. Terafab, a joint venture between Tesla, SpaceX, and xAI, was announced on March 21, 2026, and targets 100–200 billion specialized chips per year at full scale. If it gets there, it could break Tesla's looming chip supply bottleneck and reshape the economics of self-driving cars and humanoid robots. Whether it actually will is a different question.
The plan
Terafab is planned for the Giga Texas North Campus in Austin, with an initial investment of around $55 billion and a total projected spend of $119 billion across all phases, per Wikipedia: Terafab. That rivals the combined US chip investments from TSMC and Intel. Production starts at 100,000 wafer starts per month, scaling eventually to one million. The process node is 2nm — competitive with TSMC's leading edge — and Intel joined as manufacturing partner on April 7, 2026, contributing its 18A process technology.
The vertical integration is the unusual part. Design, lithography, fabrication, memory, packaging, and testing would all sit under one roof, letting engineers tweak chip architecture far faster than a conventional foundry relationship allows.
TERAFAB IS GOING TO BE INSANE.
— Lacey (@LaceyPresley) May 29, 2026
We’re targeting 100–200 billion custom AI + memory chips per year at full ramp that’s 1 terawatt (1,000 GW) of annual AI compute capacity. Roughly 50x current global AI chip output.
This isn’t incremental. This is the kind of scale that actually… pic.twitter.com/rMwj6DqsmK
Space chips first, robots second
Eighty percent of Terafab's planned output is earmarked for radiation-hardened D3 chips destined for SpaceX's orbital satellite data centers — servers in space, powered by solar energy. The remaining 20% covers AI5 and AI6 processors for Tesla's Full Self-Driving system, the Cybercab robotaxi, and Optimus humanoid robots.
Small-batch AI5 production is targeted for late 2026, with volume production in 2027. Tesla has signaled chip supply constraints arriving within three to four years, so the timing matters — Waymo and Zoox are already expanding, and a delayed ramp hands them more runway.
The credibility gap
The timelines are aggressive, and the industry knows it. Electrek's analysis notes that Morgan Stanley projects meaningful output between 2028 and 2030 — well past Musk's stated targets. Intel's 18A process yield currently sits around 65%, and the entire Terafab ramp depends on that improving substantially. The $25 billion initial build is a big bet on a foundry turnaround that Wall Street analysts are not fully buying.
Terafab's closed-loop model also means no disclosed roadmap for external sales — chips go to Tesla, SpaceX, and xAI first. Anyone else hoping to source Terafab silicon will need to wait for a strategy that hasn't been announced yet.