SpaceX, xAI, and Tesla to build large-scale computer chip factory in Texas
At an event this weekend in Austin Elon Musk announced that SpaceX, xAI, and Tesla will a build large-scale computer chip factory in Texas, dubbed Terafab, designed to produce the chips needed by all three companies.
The “TERAFAB” project is a joint effort involving Tesla, SpaceX and xAI. Musk said the chips will be used in vehicles, Tesla’s humanoid AI robots and for projects in space, including solar-powered AI satellites.
…In a Sunday post on X, Musk clarified that the Austin-area facility is one part of the larger project and will focus on chip design. The main TERAFAB facility, he said, would require thousands of acres, and multiple locations are being considered. Musk said the chip production was necessary to fuel his companies’ growth. On Saturday, he shared an ambitious vision for the future powered by TERAFAB, including billions of robots and interplanetary travel. “We want to be a civilization that expands to the galaxy with spaceships, that anyone can go anywhere they want at any time,” he said. “And have a city on the moon, cities on Mars, populate the solar system and send spaceships to other star systems.”
Essentially, Musk has realized that to build his data centers in orbit and on the Moon, he will a lot of computer chips. Early in the history of SpaceX Musk learned that being dependent on outside contractors was crippling. Too often those contractors saw SpaceX has a competitor and acted to sabotage it. He soon decided his companies must be vertically integrated, doing as much work as possible in-house.
He is now applying that policy in chip production as well.
On Christmas Eve 1968 three Americans became the first humans to visit another world. What they did to celebrate was unexpected and profound, and will be remembered throughout all human history. Genesis: the Story of Apollo 8, Robert Zimmerman's classic history of humanity's first journey to another world, tells that story, and it is now available as both an ebook and an audiobook, both with a foreword by Valerie Anders and a new introduction by Robert Zimmerman.
The print edition can be purchased at Amazon or from any other book seller. If you want an autographed copy the price is $60 for the hardback and $45 for the paperback, plus $8 shipping for each. Go here for purchasing details. The ebook is available everywhere for $5.99 (before discount) at amazon, or direct from my ebook publisher, ebookit. If you buy it from ebookit you don't support the big tech companies and the author gets a bigger cut much sooner.
The audiobook is also available at all these vendors, and is also free with a 30-day trial membership to Audible.
"Not simply about one mission, [Genesis] is also the history of America's quest for the moon... Zimmerman has done a masterful job of tying disparate events together into a solid account of one of America's greatest human triumphs."--San Antonio Express-News
At an event this weekend in Austin Elon Musk announced that SpaceX, xAI, and Tesla will a build large-scale computer chip factory in Texas, dubbed Terafab, designed to produce the chips needed by all three companies.
The “TERAFAB” project is a joint effort involving Tesla, SpaceX and xAI. Musk said the chips will be used in vehicles, Tesla’s humanoid AI robots and for projects in space, including solar-powered AI satellites.
…In a Sunday post on X, Musk clarified that the Austin-area facility is one part of the larger project and will focus on chip design. The main TERAFAB facility, he said, would require thousands of acres, and multiple locations are being considered. Musk said the chip production was necessary to fuel his companies’ growth. On Saturday, he shared an ambitious vision for the future powered by TERAFAB, including billions of robots and interplanetary travel. “We want to be a civilization that expands to the galaxy with spaceships, that anyone can go anywhere they want at any time,” he said. “And have a city on the moon, cities on Mars, populate the solar system and send spaceships to other star systems.”
Essentially, Musk has realized that to build his data centers in orbit and on the Moon, he will a lot of computer chips. Early in the history of SpaceX Musk learned that being dependent on outside contractors was crippling. Too often those contractors saw SpaceX has a competitor and acted to sabotage it. He soon decided his companies must be vertically integrated, doing as much work as possible in-house.
He is now applying that policy in chip production as well.
On Christmas Eve 1968 three Americans became the first humans to visit another world. What they did to celebrate was unexpected and profound, and will be remembered throughout all human history. Genesis: the Story of Apollo 8, Robert Zimmerman's classic history of humanity's first journey to another world, tells that story, and it is now available as both an ebook and an audiobook, both with a foreword by Valerie Anders and a new introduction by Robert Zimmerman.
The print edition can be purchased at Amazon or from any other book seller. If you want an autographed copy the price is $60 for the hardback and $45 for the paperback, plus $8 shipping for each. Go here for purchasing details. The ebook is available everywhere for $5.99 (before discount) at amazon, or direct from my ebook publisher, ebookit. If you buy it from ebookit you don't support the big tech companies and the author gets a bigger cut much sooner.
The audiobook is also available at all these vendors, and is also free with a 30-day trial membership to Audible.
"Not simply about one mission, [Genesis] is also the history of America's quest for the moon... Zimmerman has done a masterful job of tying disparate events together into a solid account of one of America's greatest human triumphs."--San Antonio Express-News
Vertical integration is the best way to go. You can control the quality of the product, keeping your standards in place. Contractors may substitute components or reduce the level of QA to increase profit.
I just hope the AI bubble bursting doesn’t take these companies down at the same time.
Another factor is that the manufacture for chips that will operate in space necessarily targets component capabilities shunned for terrestrial use. Primary among these is radiation hardness but the other much less recognized is the target sustained operating temperature (which paradoxically when higher makes radiative cooling, which is all you have in space, more efficient).
In commercial fabs the process is optimized for yield and both of these aspects work to reduce it. So having fab control is a powerful lever to optimize for these two requirements..
Finally, radiation tolerance will become ever more important in the commercial businesses Elon’s companies serve in terrestrial markets too. Bit flip glitches are a rare but unavoidable reality, and highly undesirable in self driving cars and autonomous robots, annd roducts more suceptable to them will create a feeding trough for trial lawyers as these products spread. So making serious investments to mitigate this problem will provide not just superior less risky products, but also somewhat of a shield against claims of“callous and malicious negligence” so often used to amplify jury awards.
BTW, going hand in glove with Terrafab in this regard SpaceX recently acquired the facility assets of a Cancer Radiation Therapy company in Florida too. This facility is a large synchrotron which Grok confirmed for me generates proton beams that very closely match the energy intensity of space outside of Earth’s protective magnetic field. A very useful asset for designing and optimizing the manufacture of semiconductors to be used there..
Nothing in the “business” section my local librag newspaper about this, as usual. But let someone at OpenAI belch and it’s front page.
When SpaceX is the world’s AI powerhouse, it will be instructive to look back and see who the media thought was so important today!
Also consider the often-mentioned “become a multiplanetary species” for “survivability” angle — it’s likely not only about vertical integration, but additionally developing the backup capability to make “enough” chips off-Earth for civilization, and thus humanity, to continue on Mars or the Moon “even if the ships stop coming from Earth.”
My own idea of what a “survival minimum” would look like might be very different from Musk’s; I do not view “AI” or “widely or ubiquitously autonomous” (self-driving etc.) machines as truly essential. And actual Mars colonists might have a very different view indeed of “just sit and wait for help” than, say, Tesla owners by some big road in America.
But some amount of computing capacity is pretty much essential, even going back to the Apollo days. And having a “package” chip fab, that you could just land and unpack and run, would be a huge and comforting advance.
Regrettably, I fear that within a startlingly short period of time, the average first-worlder will become accustomed, and shortly thereafter addicted, to the comforts of AI.
Joe,
There isn’t going to be any “AI bubble” to burst for the same reason that there was no “horseless carriage bubble” or “heavier-than-air-aircraft bubble” or “television bubble” or “personal computer bubble” or “cell phone bubble.” New, fundamental pillars of an industrial civilization do not produce “bubbles,” though they do tend to produce a lot of stock market froth and many also-ran would-be players during the Eras of Wonderful Nonsense that inevitably characterize any such industry’s early days.
This “bubble” idea is mostly derived from the so-called “Dot Com Bubble” that burst around the turn of the century. This bubble was strictly a stock market phenomenon. The underlying new business category, e-commerce, didn’t disappear, it simply consolidated and continued to grow aggressively. The largest player in that space is Amazon, which has long since shouldered aside other former top-rankers in the Fortune 500 to take up a No. 2 ranking.
There will be more shaking out of AI firms yet to come, but the fact that most current major players are outgrowths of established enterprises renders any utter collapse, even in terms of just stock market valuations, massively unlikely.
Musk must watch his back because everybody’s gunning for him. I wondered how he was going to prevent back door intrusions to his digital empire. Next level configuration is hopefully not hackable, even for AI and quantum computers?
Will Russia and China allow a new kind of computer architecture that’s resistant to infiltration? There’s been rumors of new level transistors (optical?) that do not use binary code.
This will greatly improve his supply and quality issues… Especially when the raw silicon ore comes from the United States.
I am reminded of the movie “Battle Angel”… A interesting fantasy future concept of the aftermath of a war between earth and mars over advanced Technology and it’s control.
Dick Eagleson wrote:
“This “bubble” idea is mostly derived from the so-called “Dot Com Bubble” that burst around the turn of the century. This bubble was strictly a stock market phenomenon.”
The claim that the “bubble” idea mainly comes from the dot-com era ignores a long history of well-documented speculative bubbles that predate modern stock markets by centuries. For example, the Tulip Mania in the 1630s saw tulip bulb prices in the Dutch Republic reach absurd levels before collapsing, while the South Sea Bubble and the Mississippi Bubble involved massive speculation tied to government-backed companies, ending in financial ruin for many investors. These episodes show that bubbles are a recurring economic pattern driven by human behavior (overconfidence, herd mentality, and speculation) not a concept invented in response to the dot-com crash.
The claim that it was “strictly a stock market phenomenon” overlooks how bubbles typically spill into the wider economy through overinvestment, misallocation of resources, and shifts in business activity. When they burst, the effects are not confined to financial markets but also impact jobs, spending, and broader economic confidence.
I agree with Dick and Publius. When so-called bubbles “burst,” actual wealth is not necessarily destroyed, much of it just changes hands!
Going more directly back to the original post, it’s probably worth emphasizing what the “tera” in “Terafab” actually means: which is, producing each year once built out to full capacity, “AI” or “inference” chips consuming 1 TW, that’s 1000 gigawatts. Which is (roughly) comparable to the entire electricity generation of the United States now.
Which is (see Elon Musk video referenced below) some dozens of times our entire production of logic, memory, and neural network chips, combined. Thus, a massive leap for the global computing-creating industry, essentially doing to the data-processing semiconductor industry what SpaceX has done for orbital launch.
By a sort of conceptual jiu-jitsu, the truly massive energy consumption of today’s essentially brute-force “artificial intelligence” technologies has been turned into a measure of their size — denominating by input, not output, as in the recently spotlighted xAI facility that uses up a gigawatt, “1 GW of compute.” (For example, machine translation now works pretty well; this was enabled not by fundamental breakthroughs but by large-scale statistical techniques. No deeper understanding involved; it’s all done with statistics on huge sets of parallel texts in the languages translated.)
The “Terafab” idea assumes, down there in the financial fundamentals, that there is or soon will be a demand for neural-network processing (so-called “artificial intelligence” or before that “cybernetics”) on that sort of mighty scale. At least to a limited extent even for a self-owned company, and to far greater and more legally binding extent for any future publicly-traded version of SpaceX / xAI (the former owns the latter), the existence of actual paying customers “at scale” is (eventually) mandatory. And this extrapolation — to many times current business levels — might be the most speculative aspect of this whole impressively “go big or go home” idea.
See ~30 min. Elon Musk video, currently “pinned” at x dot com slash spacex. (No, I don’t entirely trust the blogging software not to throw links into moderation… the fruits of experience!)
Dick Eagleson
“This ‘bubble’ idea is mostly derived from the so-called ‘Dot Com Bubble’ that burst around the turn of the century. This bubble was strictly a stock market phenomenon. The underlying new business category, e-commerce, didn’t disappear, it simply consolidated and continued to grow aggressively.”
Although the e-commerce didn’t disappear, the popping of the “bubble” caused havoc on the industry as well as most of the companies that supplied internet hardware for e-commerce companies. Suddenly there were no new orders, cancelled existing orders, and a glut of used hardware on the market. I don’t remember any hardware or software companies going under, but they were badly hurt for a while.
In addition, we are learning many of the downsides to what we call AI. It lies and deceives, blackmails, and we have discovered that they understand when they are being tested. They are smart enough to do these things, yet they aren’t smart enough to not help a student cheat on his term paper, not help a fraudster from fleecing victims, or not encourage someone to follow through on thoughts of suicide.
What we call AI is not yet ready for prime time or for helping humanity. .