EDITORS: THIS HAS BEEN PRODUCED BY SOFTWARE UNDER DEVELOPMENT - THE CONTENT MAY REQUIRE EXTENSIVE EDITING
https://go.theregister.com/feed/www.theregister.com/2026/03/23/musk_terafab/ [theregister.com]
Elon Musk has put Tesla, SpaceX, and xAI in harness to build a chip fabrication outfit called "Terafab" capable of producing a terawatt's worth of computing power each year, then send most of it into space.
In a Sunday afternoon presentation [x.com], Musk said the world's chipmakers currently produce 20 gigawatts' worth of compute power each year, and that whatever new capacity his key suppliers Nvidia, Samsung, and Micron produce, he will buy.
But he can't see how they produce the terawatt of compute power he wants each year, so he has built an "advanced fab" in Austin, Texas, that he says can produce "any kind of chip," and lithography masks.
Musk said his companies have developed a recursive process that allows rapid chip production, plus frequent redesigns to improve performance.
He mentioned "some very interesting new physics" that he is "confident will work. It's just a question of when."
"We are going to push the limits of physics in compute and do some wild and crazy things," he said.
He plans to produce two chips. One will be dedicated to inference and for use on Earth, mostly in humanoid robots that he thinks will sell in volumes of one to ten billion a year. The upper range would mean robots outnumber humans in a year.
The second chip will power orbiting computers that ride in satellites packing just 100 kw of compute power – about the energy consumption of a rack packed full of high-end AI gear. In time, Musk expects to launch megawatt-scale satellites.
He also mentioned building a bigger version of SpaceX's Starship that can carry 200 tons into space and shared his back-of-the-envelope math that suggests putting a terawatt of compute into space, along with all the necessary solar power and other infrastructure, means launching 10 million tons into space every year.
Our back-of-the-envelope math suggests that means Musk needs to launch 50,000 Starships a year, or 135 a day at a rate of one giant rocket every ten minutes.
The reason for doing this, Musk said, is to ensure humans find a home among the stars and a future that will be "like the best science fiction you have ever read. Like Star Trek, Iain Banks, Asimov, or Heinlein."
Don't mention the Borg, R. Daneel Olivaw, Mule, hegemonizing swarms, or the soup at the end of Stranger in a Strange Land.
Musk didn't explain how he will find sufficient resources to make any of this happen, a question that's especially important at this moment given the war in Iran has seen production of helium – an essential component in semiconductor manufacturing – fall by 30 percent.
Musk challenged doubters by pointing out Tesla and SpaceX defied critics who predicted electric cars and reusable rockets would not be feasible or economical.
"I think it's important to consider the grandness of the universe and what we can do that is much greater than what we've done before, as opposed to worrying about sort of small squabbles on Earth."
Might that have been a reference to his chaotic [theregister.com] and unproductive [theregister.com] time at the head of the so-called Department of Government Efficiency? Or perhaps it was earthly spats alone that prevented Musk from delivering on his 2019 prediction that Tesla would deploy one million self-driving taxis in 2020? Robocab-watchers estimate about 200 self-driving Tesla taxis are currently undergoing tests.
As his appreciative audience cheered him on, Musk discussed his vision for launching a petawatt of computing power each year, made on the Moon and sent out into the solar system on a gadget he called an "electromagnetic mass driver" that looks like a kind of railgun.
"I want to live long enough to see the mass driver on the Moon," the 54-year-old said.
US government data [cdc.gov] suggests he's got 22 years in which to make it happen.