Forgot your password?
Close
typodupeerror
Hardware

Elon Musk Announces $20B 'Terafab' Chip Plant in Texas To Supply His Companies (yahoo.com) 126

Posted by EditorDavid from the chipping-away dept.
"Billionaire Elon Musk has announced plans to build a $20 billion chip plant in Austin, Texas" reports a local news station: Musk announced on Saturday night during a livestream on his social media platform X that the plant, called "Terafab," will be built near Tesla's campus and gigafactory in eastern Travis County. The long-anticipated project is a joint venture between Musk-owned properties Tesla, SpaceX and xAI... The Terafab plant is expected to begin production in 2027.
Musk "has said the semiconductor industry is moving too slow to keep up with the supply of chips he expects to need," writes Bloomberg — quoting Musk as saying "We either build the Terafab or we don't have the chips, and we need the chips, so we build the Terafab." Musk detailed some specific plans, including producing chips that can support 100 to 200 gigawatts a year of computing power on Earth, and chips that can support a terawatt in space, but gave no timelines for the facility or its output... The facility is expected to make two types of chips, one of which will be optimized for edge and inference, primarily for his vehicle, robotaxi and Optimus humanoid robots. The other will be a high-power chip, designed for space that could be used by SpaceX and xAI... Musk said he expects xAI to use the vast majority of the chips.

During the presentation, Musk also unveiled a speculative rendering of a future "mini" AI data center satellite, one piece of a much larger satellite system that he wants SpaceX to build to do complex computing in space. In January, SpaceX requested a license from the Federal Communications Commission to launch one million data center satellites into orbit around Earth. Musk said that the mini satellite he revealed would have the capacity for 100 kilowatts of power. "We expect future satellites to probably go to the megawatt range," Musk said.

Raising money to build and launch AI data centers in space is one of the driving forces behind SpaceX's planned IPO later this year. SpaceX is expected to raise as much as $50 billion in a record-setting IPO this summer which could value it at more than $1.75 trillion, Bloomberg News reported earlier.
This discussion has been archived. No new comments can be posted.

Elon Musk Announces $20B 'Terafab' Chip Plant in Texas To Supply His Companies More

Elon Musk Announces $20B 'Terafab' Chip Plant in Texas To Supply His Companies

Comments Filter:

  • Guarantees (Score:5, Insightful)

    by algaeman ( 600564 ) on Sunday March 22, 2026 @05:57PM (#66054844)
    Make sure the plant is built and running before they get a penny of the tax breaks

    • Re:Guarantees (Score:5, Insightful)

      by Kisai ( 213879 ) on Sunday March 22, 2026 @07:02PM (#66054958)

      Remember "full self driving" , remember "hyperloop"

      Yeah, this is more of that.

    • Politicians already have their tonsils on his nuts.

    • I'm not worried about the tax breaks (Score:4, Insightful)

      by rsilvergun ( 571051 ) on Sunday March 22, 2026 @08:13PM (#66055080)
      I'm worried about yet another pump and dump. Like when Elon said you would have self-driving cars 10 years ago or when he said he'd have a $40,000 electric truck or like when he said he would have cheap electric semi trucks with super range.

      Over and over and over again he makes the kind of promises that investors use to make decisions and he's clearly lying but he somehow gets away with it. Traditionally doing something like that would be considered investment fraud. But he hasn't cost anyone any real money. Yet.

      Oh and what's not forget he used his AI company linked with Tesla and SpaceX to buy up all the bad debt he had from Twitter... That is an enormous conflict of interest that should see him in prison already...

      • Over and over and over again he makes the kind of promises that investors use to make decisions and he's clearly lying but he somehow gets away with it. Traditionally doing something like that would be considered investment fraud. But he hasn't cost anyone any real money. Yet.

        Lots of promises made and not kept, yet nobody pulling him up on it (from his own party at least)? Sounds like a description of Trump to me...

      • Re:I'm not worried about the tax breaks (Score:4, Informative)

        by AmiMoJo ( 196126 ) on Monday March 23, 2026 @05:45AM (#66055644) Homepage Journal

        Speaking of which, a jury just decided that he has to compensate Twitter investors for his lies when he was looking to buy it. Could be as much as $2.5 billion.

        2026 will be the 10th anniversary of Musk starting to sell Full Self Driving as a "coming in the next few months" feature. The original promise was that it would drive you to work while you browsed your phone, drop you off and go park itself. I'd be asking for a refund at this point.

  • I'm sure it'll be ready- (Score:4, Insightful)

    by locater16 ( 2326718 ) on Sunday March 22, 2026 @06:05PM (#66054854)
    -next year, for the next 10 years.

  • 1 . California? Not Musk's first choice. They'll mostly vote Democrat. All that will do is swell the democratic base in Austin. You're also going to have a tough time getting [most] Californians to move to Texas unless they like hot sticky weather and the Republican values.

    2. H-1B? Not with Trumps $100,000.00 visa fee. But Musk's upside to this is that they can't vote.

    3. No engineers or white collar employees, just blue collar workers and AI. (Musk's wet dream). Blue collar workers are for the most part R

    • Are modern fabs really labor intensive these days?

      • Re: (Score:2)

        by hwstar ( 35834 )

        Robots can't solve technical issues on the production line. AI might however.

        • Precisely. I have seen Asianometry videos on YouTube about how fabs work - the videos were about Intel's and TSMC's fabs in Arizona. A lot of it involved people who had to change in and out of those suits every time they entered or left a facility. Reading the comments there, it looked like a lot of those jobs were very specialized, dead-end jobs, which were only secure as long as the fabs ran

          So it would seem that a lot of the operations could be automated further like on an assembly line, so that the

      • Yes because making chips is not about turning machines on and everything magically works. At the leading edge of chip fabrication is engineering to determine how to make things work and at acceptable yields.

        • That was true about older fabs at larger process nodes. Is it still true when the process nodes are below 10nm, which is now approaching atomic-level territory? It would seem that this would be one of the jobs that AI would do better

          • Re:Where's the talent going to come from? (Score:4)

            by UnknowingFool ( 672806 ) on Sunday March 22, 2026 @07:22PM (#66054986)

            That was true about older fabs at larger process nodes. Is it still true when the process nodes are below 10nm, which is now approaching atomic-level territory? It would seem that this would be one of the jobs that AI would do better

            Bahahahahaha. And how would AI fabricate chips better when it gets details wrong all the time. Chip fabrication is intensively detail oriented. I can use see it now: "AI thinks we should use carbon as the substrate instead of silicon."

              • Please show me a non LLM based AI model that can solve the hard engineering issues that is at the leading edge of chip fabrication. TSMC, Intel, and Samsung would like to know.
                • AI and engineering are not my areas of expertise but there are certainly categories of AI that are highly valuable to some types of R&D. My point was more that LLM is the category of AI that "gets details wrong all the time".
                  • You said AI is not LLM. When challenged for an AI model that can solve the hard engineering problems at the leading edge of chip fabrication, now you are responding that you do not know. In other words, you don’t know but you are sure you are right.
                    • Back to the point, which AI model would help the engineers at TSMC, Samsung, and Intel solve the hard engineering problems with chip fabrication at the leading edge? If you know, you should contact them with your suggestions. If you don't know, you can't assume that AI will solve their problems. AI with the right expertise can help remove a lot of trivial work. The issue with AI right now is people who have no expertise thinks it can solve any problem in a field which they have no expertise. It is a classi

        • Can't wait to see how many H1-Bs he hires because it's cheaper.

      • Re:Where's the talent going to come from? (Score:5, Interesting)

        by coofercat ( 719737 ) on Monday March 23, 2026 @10:56AM (#66056212) Homepage Journal

        It's been over 20 years since I was last in a fab, but...

        The actual making of wafers is pretty much automated. Apart from loading up materials and such like, it's pretty much hands-off.

        However, let's say you've got a brand new production line available. You've got a room full of designers who make up a silicon design. Once that's done (and simulated, etc), it starts being a bit of a dark art. It's not like just sending it off to a laser printer and waiting for it to come out. What tends to happen if you get various 'artefacts' in the wafer, which are a symptom of the specific design you chose and the finer details of the manufacturing process, the input materials, etc. Some of those things are fixable in the manufacture, some of them need to be designed out. Then you get a few wafers which look good, so you go to full manufacture. Then you find weird problems where the yield of the devices you're making drops off a bit, so you're back to tweaking the manufacture (because by then, tweaking the design means a whole load of development/testing etc)... and so on.

        So in answer to your question... the clean room isn't exactly packed with semiconductor engineers and machine experts, but there do have to be quite a few of them to work out how to make the technology make the thing you want to the level of quality you can accept. They'll be predominantly doing that at the start of a new product line, or a new machine/process, but they're still going to be doing some amount of work throughout the entire time you're manufacturing that same old chip you've been doing for the last 20 years.

    • Re:Where's the talent going to come from? (Score:5, Informative)

      by dfghjk ( 711126 ) on Sunday March 22, 2026 @08:19PM (#66055098)

      It's hysterical you think the only sources of "talent" are California and H1-B. Also, apparently you are unaware that Austin is heavily Democratic, just how much "swelling" do you think will happen in a major metropolitan area? And where do you think the IBM Power architecture came from? Motorola? Freescale? All Austin. You are unbelievably ignorant.

  • What's the backlog at ASML? (Score:3)

    by ebunga ( 95613 ) on Sunday March 22, 2026 @06:19PM (#66054884)

    Because a giant building with no equipment isn't exactly useful.

    • That’s what his company xAI is for. I am sure they are working on this problem right now. People just have to believe.
    • Last I knew, about 18-24 months, and growing. With enough money you might be able able to buy someone else's earlier delivery spot, but that is not going to be more than drop in the bucket into the claimed needs of a terafab.
    • They have backlogs? I'd have thought that given the cost of equipment, those companies would have an uphill struggle finding customers, and keeping the production lines busy

    • Re: (Score:3)

      by gweihir ( 88907 )

      That and would ASML even have capacity to sell to him? Somehow I doubt that. Building such a fab is a 10 year undertaking. If you have experience with it. There is no "move-fast-and-break-things" in this space.

    • It's not ASML that'll be the problem. They'll buy a system and it'll be there in crates ready to be isntalled long before:

      A) the permits are in place. This is Texas, so maybe "faster than California", but it'll cost.

      B) The water and power is secured; again it might go faster but it's 18-24 months. However,

      C) This is Texas, the power grid has been unreliable. Since foundries run 24/7 fully automated to produce the kinds of chips he's talking about, they'll need reliable backup power. Backup gener

      • For power, he will likely divert a bunch of solar panels and grid-scale batteries from Tesla.

        The bigger issue is that he wants to put this close to sources of vibration, like the Tesla gigafactory that uses high impact tools to shape metal. Apparently reputable commenters elsewhere have said that these impacts, while invisible to human sensations, are likely enough to affect high-sensitivity chip manufacturing operations. Existing fabs all over the world have to take into account traffic from nearby highway

        • Re: (Score:2)

          by tragedy ( 27079 )

          He was good as an idea man for a while, but his ideas have lost contact with reality.

          Depends on how you define "idea man". He was good there for a while at identifying ideas from the past whose time had come. Stuff off the covers of popular mechanics, out of science fiction, that sort of thing. So if an idea man who picks pre-existing ideas and tries them again, then he's an idea man. Early on, I appreciated that, before learning more about him (first big red flag was maybe 2005 or so when I learned that he was the kind of boss who fires people on the spot for disagreeing with him/making a

  • C'mon Elon, you could have used petamicrowatt units to justify calling it a petafab chip plant. I'm disappointed in your imagination for exaggeration

  • TSMC has had all kinds of problems getting it's US fab going (and that doesn't even do the entire process of making a finished chip) and Intel with its decades of experience in chipmaking has had many problems with its newest stuff.

    It would be great to see Musk build his own chips but it's not as simple as buying a bunch of stuff from the ASML catalog...

    • Re: (Score:2)

      by dfghjk ( 711126 )

      ...and "his own chips" is an interesting topic as well. What is he going to build? Is Elon Musk now ALSO the world's foremost chip designer?

      Buying all your talent from competitors MAY give you the ability to produce something but it does NOT contribute to advances in the industry, only to who owns them. That's the best possible outcome here, that Musk doesn't fuck it up entirely and just loots from others in the industry. That's all he ever achieves, half the time even that fails.

      • Is Elon Musk now ALSO the world's foremost chip designer?

        I'm sure that he'll insist on having that sort of title within the company, and that his employees will be obligated to support that in any public comments

        That's assuming this fab actually gets off the ground. If it doesn't, it will be someone else's fault.

        • Doesn't the FSD hardware already use custom chips?

          I assume they'll be making a new chip (at least to reduce production costs by shrinking) based on their current custom chip.

          I'm skeptical they'll get the plant up and running though with the history of doing such on the US being sketch overall.

          I'm also skeptical that they'll really benefit from it, though I can see why it's compelling to bring in house watching supply chains being strained right now.

    • What lithographies would Tesla or SpaceX be looking at? It can't be the 10nm-2nm ones: those would never be stable at automotive or space temperatures. I'd imagine they'd be looking at processes w/ much larger nodes, to allow for those temperature ranges

      In which case, he doesn't need to procure the latest & greatest equipment: he can get time-tested equipment from those vendors

  • You can readily find articles (Score:4, Informative)

    by rsilvergun ( 571051 ) on Sunday March 22, 2026 @07:27PM (#66054994)
    explaining why this is nonsense. Never mind the fact that 20 billion isn't going to build you the kind of Fab needed to build the kind of chips he needs we will just ignore all that. You can find articles where they talk about how musk just does not have the chip design people that Nvidia and Apple and Google have.

    This is more pump and dump. Trump has changed the regulations regarding what can go into your 401k and they're going to dump musk's companies stock into them and let it collapse taking your retirement with them. If you are old enough to have a pension you will survive that but otherwise prepare to be flat broke in your old age no matter how hard you work or how much money you make.

    If you don't believe me go look up the SpaceX IPO video Patrick Boyle did on youtube. The too long didn't watch is SpaceX has already maxed out all the launch customers that are available so the only growth opportunity is his internet satellites and he's already maxed out the number of customers for those.

    They are about to dump 1.5 to 2 trillion dollars of bad stock into your 401k and there's practically nothing you can do about it.

    • Re: (Score:2)

      by gweihir ( 88907 )

      In theory, $20B is just about enough to finance such a fab. But it does not get you the people or the experience or the organization to make it work. This is not even a problem of money. He would have to hire senior people away from TMCS or Samsung and I doubt any of them would even contemplate coming to the US at this time. Oh, and the $20B is just the fab. A fab does not come with chip-design capabilities, that is not its job.

      So, yes, this is stock-market manipulation, plain and simple. One of the things

      • Both of them wanted at least 60 billion to build the kind of Fab that could make the processors that could do AI shit. Now I'm sure that number is inflated but let's say it's half what Intel and AMD claim, musk still needs another 10 billion. That's before we count the inevitable cost overruns.

        But yeah you are correct he does not have people with the expertise needed to make the chips that that kind of a Fab would make. And those people are so rare they are monopolized by intel, AMD and Nvidia. There ar

        • Re: (Score:2)

          by gweihir ( 88907 )

          Indeed. It is just problems on all fronts.

          Now Musk could try to buy, say, TMSC and Intel, but, funny thing, he probably does not have enough cash for than and the current owners will very likely not want to sell anyways.

        • Re: (Score:3)

          by ceoyoyo ( 59147 )

          You can "do AI shit" on a processor you can build in your mom's basement, or on a 1970s Z80 that's already there. Tesla wanted to do it a bit faster than that with less power so their HW3 was built by Samsung on a 14 nm process using DUV lithography. The HW4 is probably a TSMC 7 nm which could be DUV or EUV lithography.

          Intel announced their first 14 nm fab in 2011 at $5 billion and they quote their 7 nm fabs at $10 billion.

          Tesla/SpaceX might be talking about a cutting edge fab, in which case $20 billion is

  • A millions data Centre satellites pumped up into the space. This is just satire at this point, no?
  • Unless they are referring to the rate at which electricity production capacity is being increased (which theyâ(TM)re not), "gigawatts a year" makes no sense.

  • There's no way on (or off) Jupiter's Green Earth that you could ever put a terawatt of anything in space. Every one of those watts would have to be radiated into space somehow. Can you imagine how large the heat sinks would have to be? Keep in mind that radiating heat in a vacuum is about the most inefficient way to get rid of heat that one can think of. On Earth, you can use air and water to move heat around and then dump the air or heat into the environment. In space, you only have radiation, which is bas

    • Around a billion m2 of heatsink, it's a little much.

      The heatsink can be very light though, they need no real structural strength beyond being able to contain the pressure of the refrigerant (say ethanol at around 1 bar in lots of parallel channels in a thin sheet). The PV can be even thinner.

  • First, building such a fab, if you know what you are doing and have respective experience, is a 10 year or longer thing. Second, you need people with very specific and very rare skills to build and then operate it. I would be surprised if Musk can even hire those. He would have to hire them away from TMCS or Samsung and these people are likely not even interested in moving to the US, let along working for Musk. Third, you need the equipment. This is exceptionally expensive and there are long waiting times.

  • Pollution incoming (Score:3)

    by sinkskinkshrieks ( 6952954 ) on Monday March 23, 2026 @02:11AM (#66055446)
    Santa Clara County, California has more Superfund-registered hazardous waste sites than any other American county because of previous generations of chip manufacturing fabs and suppliers. Austin will repeat history if it doesn't closely regulate and monitor this if it is ever built.

    • Re: (Score:2)

      by dgatwood ( 11270 )

      Santa Clara County, California has more Superfund-registered hazardous waste sites than any other American county because of previous generations of chip manufacturing fabs and suppliers. Austin will repeat history if it doesn't closely regulate and monitor this if it is ever built.

      Yeah, that's exactly what I was thinking. Texas having weak environmental laws is presumably a big part of why Musk is doing it there.

  • I've only heard of 4 fabs and two of them died already.
  • All kidding aside, does anyone trust him?

  • I don't see how this fab can have adequate scale just supplying Musk. It will be a giant money sink.

    • How many types of chips go into a car? How hard might it be to reduce a number of them to one of a few categories - CPUs, NPUs, Memory, ASICs, DSPs, FPGAs,... So that they can rake up the volumes in each of them?

  • And honestly, it is closer to the latter.

    $20 billion sounds like a lot until you compare it to the reality of building a modern semiconductor fab. TSMC has already spent several times that on its Arizona fab, and it is still not operating at full capacity.

    Why? Because the hard part is not just construction. It is the supply chain.

    Take water, for example. Semiconductor manufacturing requires ultra pure water, but water chemistry varies by location. In Phoenix, TSMC had to design and build a bespoke purificat

Slashdot Top Deals

What this country needs is a good five cent nickel.

Close