4 minMusk unveils Terafab in Austin: Tesla, SpaceX and xAI will build their own chip factory, aiming for a terawatt of compute a year
At a livestreamed keynote in Austin, Elon Musk announced Terafab, a joint chip-making project of Tesla, SpaceX and xAI. It starts with an 'advanced technology fab' that can make logic and memory chips and their masks under one roof, and makes two kinds of chips: one for Optimus and cars, one for AI in space.
Elon Musk has announced what he called "the most epic chip building exercise in history by far". At a livestreamed keynote in downtown Austin on Saturday evening, March 21, he presented Terafab: a joint project of Tesla, SpaceX and xAI to make their own AI chips, at a scale far beyond what today's chip industry produces.
What Musk announced
- A first fab in Austin: the project starts with an "advanced technology fab" in Austin. Musk thanked Texas governor Greg Abbott, who was in the audience, for the state's support. An exact site was not given.
- Everything under one roof: according to Musk, this fab will have the equipment to make chips of any kind, logic and memory, plus the equipment to make the lithography masks. Packaging and testing happen in the same building. That allows a fast loop: make a mask, make a chip, test it, change the design and repeat. Musk said he knows of no other place in the world that can do all of this in one building, and claimed the speed of improvement will be about ten times better than anywhere else.
- Two kinds of chips: one optimised for running AI on the device itself, mainly for Optimus robots and Tesla cars, and one high-power chip designed for space. The space chip has to cope with radiation and charge build-up, and is meant to run hotter than chips on Earth so that the satellites need smaller radiators.
- New physics: beyond conventional chips, Musk said the team will try "wild and crazy things", including new physics he expects to work eventually.
Why Musk says he needs it
The goal is a terawatt of AI compute per year. According to Musk, the world currently produces about 20 gigawatts of AI compute a year, and all existing chip factories combined deliver only about 2 percent of what his companies want. He stressed that Tesla, SpaceX and xAI remain grateful customers of Samsung, TSMC and Micron, and will buy everything they can make, but those suppliers do not want to expand as fast as he would like. "We either build the Terafab or we don't have the chips," he said. "And we need the chips, so we're going to build Terafab."
What it means for Tesla and Optimus
For Tesla, the first chip type matters most. Musk expects humanoid robots to be built in far larger numbers than cars eventually: somewhere between one billion and ten billion a year, against roughly 100 million cars worldwide today, and he wants Tesla to make a large share of them. Each of those robots needs an AI chip. Musk also pointed to what the companies have done before: Tesla now builds about 2 million electric cars a year, xAI built what he called the first gigawatt-scale compute cluster, and SpaceX has landed rockets more than 500 times.
The space part
Most of the compute, Musk said, will end up in space: perhaps 100 to 200 gigawatts a year of chips on Earth and around a terawatt in orbit, because power on Earth is limited. In space the sun always shines, solar panels deliver at least five times more energy, and they need no heavy glass or frames. He showed a small "mini" AI satellite of about 100 kilowatts next to a Starship V3, and said later satellites will move to the megawatt range. Reaching a terawatt would require about 10 million tons to orbit a year, which he said Starship can achieve, with payload rising from about 100 to 200 tons per launch. He predicted that AI in space could become cheaper than AI on Earth within two or three years.
Musk went even further: after a terawatt, he talked about an electromagnetic mass driver on the Moon, operated with Optimus robots and people, to launch compute into deep space without rockets.
The open questions
- No timeline or cost: the keynote gave no completion date and no budget. Reports put the cost of the first phase at around $20 billion, but that was not confirmed on stage.
- A very hard industry: advanced chip factories are among the most complex projects in the world, and even experienced chipmakers struggle with delays.
- Ambition versus track record: many of the numbers, from billions of robots to a terawatt in orbit, are long-term goals, and several of Musk's earlier timelines have slipped.
What is clear is that Tesla, SpaceX and xAI now openly plan to share one chip supply chain, a step that ties Tesla's robots and cars more closely to Musk's space and AI companies.
Sources (3)
Images are illustrations; some are made with AI.

















