AUSTIN, Texas — Elon Musk used the Baron Investment Conference on Friday to spell out the scale of Tesla's most ambitious manufacturing bet yet: a semiconductor operation, internally dubbed "Terafab," aimed at eventually producing a full terawatt of AI chips every year.
The target is deliberately staggering. Musk estimated that the rest of the world will produce on the order of 20 gigawatts of AI chips this year, rising to perhaps 35 to 45 gigawatts in 2027. A terawatt of annual output would be roughly 30 times that global production — inside a single company's plan. "It's terrifyingly large," Musk said, noting that the Terafab name is a nod to terawatts rather than to the planet.
A Research Fab Feeding a Giant Production Fab
Musk laid out a two-site structure. A research fab at Gigafactory Texas in Austin would develop and prove new manufacturing processes, and the ones that work would graduate to a far larger production facility planned for Grimes County, near Texas A&M. The aim is to compress the chip-development loop from a year to about a month. "Something that would normally take a year, I think we may be able to reduce to a month," he said, describing a fab designed for rapid iteration rather than one frozen around a single process.
That appetite for in-house silicon is driven by Tesla's own products. The company's push into autonomy — from Full Self-Driving to its Cybercab robotaxi — depends on an ever-growing supply of custom AI chips, and Musk has made clear he does not want that supply constrained.
Not Starting From Zero
Tesla already has silicon partners in motion. Samsung is building a fab in Texas to produce Tesla's next-generation AI5 and AI6 chips, and TSMC confirmed last month that it is in discussions about a potential Terafab role. Musk argued the strategic logic is hard to ignore: by his estimate, total U.S. chip-fabrication capacity, even counting every announced project, is only about 10 percent of Asia's. For a company betting its future on real-world AI, owning more of that stack is a hedge against scarcity.
The robotaxi effort illustrates the stakes. Tesla's autonomous fleet has been expanding testing well beyond Austin, and every additional vehicle and every software improvement adds to the inference workload the company must feed.
A Trillion-Dollar Plan, Built in Phases
Baron pressed on the price tag, suggesting a plant of that scale could cost a trillion dollars. Musk's reply was that the money would not be needed all at once, because Terafab would be built in stages, with each phase funding and informing the next. That phased approach mirrors how Tesla scaled its vehicle factories and energy business — start smaller, learn fast, then expand aggressively.
Whether Terafab reaches a full terawatt is a question for the 2030s, not next quarter. But the Baron conference made the direction unmistakable: Tesla intends to become a serious chipmaker, not just a chip buyer. Details on the company's broader AI efforts are available on Tesla's AI page, and Musk signaled the first research-fab milestones in Austin are the next thing to watch.
