3 Companies, 1 Chip Factory: Who Runs Elon Musk's Terafab?

Kind: captions Language: en [music] On Friday, October 2nd, a newsletter reported that TSMC was exploring a role in Elon Musk's chip factory. The next day, Musk replied, "Just discussions, but something may come of it." On Monday, before the market opened, Intel's stock fell more than 4%. Intel had been the only chipmaker announced as a partner in the project. No deal with TSMC has been announced. But this isn't a story about how many chips the factory can make. It's about who gets to run the room where they're made. Here is where things stand. The factory is called Terapob. Bloomberg reported that silicon manufacturing is not planned to begin until 2029. The governor of Texas has announced more than 16.8 billion of investment for its first phase. Three parties now have something at stake. Musk says he cannot buy enough chips, so he wants [music] a source of his own. Intel joined in April and its stock has more [music] than doubled since even after Monday's drop, which had narrowed to a loss of roughly 2% by late morning. Yet only about 5% of the revenue from Intel's chipmaking arm comes from outside customers. TSMC already supplies Tesla. Whether it wants more than that, it has not said. It declined to comment on the report. As reported, the story did not describe a takeover. One version has TSMC owning and running its own plant in Texas with Terapab as its main customer. Nothing like it is agreed. This video [music] sets the three sides next to each other on paper. What each has said, what each has left out. In May, SpaceX filed paperwork for its stock listing, and one passage in it says how firmly anyone is bound to this project. That passage comes soon. First, why Musk says he needs the factory. Elon Musk says his companies cannot get enough chips. He also says he is grateful to the companies [music] that make them. Both statements trace back to the same project. At Tesla's shareholder meeting in November 2025, Musk said the company might have to build a giant chip plant of its own, a terra, and raised Intel as a possible partner. The formal launch came 4 months later in Austin on March 21st. We either build the Terraab or we don't have the chips, he said. Then he named Samsung, TSMC and Micron as suppliers he remains thankful for. His argument was about scale demand from his businesses he said would eventually exceed what the whole industry can deliver. He put a number on it. Global output of AI computing power he said is about 20 gawatt a year. Everything else on Earth by his count adds up to roughly 2% of what his companies need. That's his estimate offered at a product reveal. It is a claim, not a measurement. [music] Where would the chips go? According to SpaceX, to three places. Tesla's Cyber Cab, a planned robo taxi. Tesla's Optimus, a humanoid robot, and data centers in orbit, which SpaceX says it is developing. Samsung and TSMC make Tesla's chips today. Terrafab is aimed at what comes after. The plan is to do every step under one roof. The term for it is vertical integration. Chip design, fabrication on thin slices of silicon called wafers, memory, advanced packaging and testing, all at one site. Electric reported that no company does all of it at this scale, TSMC included. Consider who would be doing it. [music] Tesla builds cars, SpaceX builds rockets, and runs an AI business. Tesla designs its own chips, and designing a chip is a different trade from making one. Bloomberg noted in April that Musk's companies had never made them, not once. [music] To judge the next promise, look at the last one. In September 2020, Tesla unveiled a new battery cell, the 4680. Musk told the audience the technology was not yet stable, and said he hoped to be in mass production by 2021. In its report for the last quarter [music] of 2025, Tesla said it now makes dry electrodes for those cells in Austin, for both the anode and the cathode, the two electrodes inside every cell. In August, Tesla said it was running the first large-scale cathode plant in North America, also in Austin. That is nearly 6 years after the unveiling. In between, doubts were public. In 2024, Robin Zen, chairman of cattle, the world's largest battery maker, told Reuters that Musk promises too much on timelines. Something that might take 5 years, Zung said. Musk says will take two. That same year, the information reported that Tesla was losing 70 to 80% of its cathodes in test production. Conventional battery makers, the report said, lose fewer than 2% of their components to defects. None of this proves anything about chips. Batteries and semiconductors are different trades, and Electric said so, too. But Electrek called the battery program Tesla's closest comparable venture. It took nearly 6 years to get there. On the chip side, the record is mixed. Tesla built a strong design team. Electric reported that Jim Keller, a well-known chip architect, joined in 2016 and left in 2018. Ganesh Venqatarammanan who led the Dojo supercomputer project left in late 2023. In August 2025, Musk shut dojo down and Peter Bannon who led custom chip design left the company. Making chips takes a different workforce. One electric said in March that Tesla had never employed and was then recruiting. Outside voices are cautious. Jensen Huang runs Nvidia, a company that depends on TSMC. In November 2025, at an event in Taiwan, he said, "Building advanced chip manufacturing is extremely hard." He also reportedly told reporters that matching TSMC is virtually impossible. Electric noted that Hang has every incentive to want more chip capacity in the world. Other outlets noted that he was in Taiwan meeting TSMC executives when he said it. So, who is actually bound to this? Doubts are one thing. Here is what binds the partners. In May, SpaceX filed its paperwork for a stock listing. In the risk section, it wrote, "Neither Tesla nor Intel are obligated to remain a part of the project." The filing says, "Terraab is meant to ease a shortage of AI chips and that it may not succeed." SpaceX also says it expects to keep buying much of its computing hardware from outside suppliers. The shortage it has in mind is particularly computing in orbit at scale, which it says needs significantly more chips than it can get today. It described its arrangement with Tesla as a framework agreement and said it may not reach a final one. Risk sections are cautious by design. They list possibilities, including unlikely ones. So this is a company warning its investors, not a forecast of failure. Still, it sets the baseline. As of May, two of Terrafab's partners were free to walk away. That is a different question from who would run the plant, but it shapes every answer to it. [music] Start with the size. The target is 1 terowatt of computing output a year. By Musk's own numbers, that is about 50 times what the world's AI chips produce today. It is the same 2% seen from the other side. Now, the money. On August 6th, Texas Governor Greg Abbott confirmed the site and announced more than 16.8 billion of investment for the first phase. SpaceX's own earlier estimate for what [music] it called the first phase was about 55 billion. The two figures may not count the same things. SpaceX's [music] estimate for the whole project is up to 119 billion. Either way, the number has moved. There are two places to keep apart. In Austin, on the north campus of Tesla's Giga Texas plant, a research fab broke ground in April. A fab is a fabrication plant, the factory where chips are made. The full production complex is planned for Grimes County on the site of a retired power station at Gibbons Creek about 20 mi east of College Station. The plan calls for a campus of about a 100 million square ft. Now, the output, the plan Terraab has laid out starts at a 100,000 wafer starts a month and grows toward a million. A wafer start is one of those silicon slices entering the line. Bloomberg reported that a smaller pilot line comes first, handling 3,000 wafers a month. The plant has four jobs all under one roof. The first is design. Tesla has done [music] this before. Its in-house team designed the chips that run autopilot. For Terraab, Musk said the complex will technically be two fabs, each making one chip design. one for Tesla's vehicles and Optimus robots, one for space, built to run hotter than a standard chip. The second job is making the wafers, and this is where the plant either works or it doesn't. The problem is printing circuits so small that a speck of dust can ruin one. The solution is a machine that prints circuits onto silicon using extremely short wavelength light. An extreme ultraviolet lithography machine. It prints one layer at a time. And a finished chip is many layers stacked and aligned. Picture drawing a city's entire street grid on the head of a pin with every line crisp. Musk has called Austin's research fab the precursor to Terapab. Intel's name comes up again here. On Tesla's April earnings call, Musk said the plan is to use Intel's next manufacturing process called 14A and that he expects it to be mature by the time Terrafab scales up. The process is not in volume production yet. In July, Intel said it is committed to a high volume ramp of 14A in 2028. The third job is memory. One report noted that making logic and memory chips in [music] one facility departs from industry norms. Musk said Terrafab will make both along with photo masks, the pattern plates used to print the circuits. The fourth job is packaging and testing. A modern chip is often several small pieces of silicon called dyes joined into one module on a chip packaging line. Each wafer is also probed on a wafer testing station before the pieces are cut apart. Trendforce reported that TSMC's most advanced packaging capacity is tight. Intel's name comes up once more. On April 7th, Intel posted that its ability to design, fabricate, and package ultraigherformance chips at scale would help Terraab. Packaging is on that list. What Intel's role will be, it hasn't said publicly. Its chief executive reportedly told staff in April that the scope of the work would be disclosed within weeks. In the reporting since, no fuller account has appeared. Now follow one wafer through all four. It starts as a polished disc of silicon and enters a clean room. A room where the air is filtered almost free [music] of dust. Electric wrote that leading edge fabs run at the cleanest classes of the international standard. The cleanest allows at most 10 particles of a tenth of a micron in a cubic [music] meter of air. The disc goes into the machine where light draws one layer. Then it goes back for the next, each aligned with the one below. That is the job the plant has to do and it can't do it without the machines. Reuters reported in May that the chief executive of the Dutch company that builds them had spoken with Musk directly and called him very serious. He also said projects like Terraab are expected to put growing pressure on suppliers capacity with order backlogs already at record levels. Money alone doesn't put a machine on the floor. No machine in the plant can answer the next question. That one belongs to the people who signed the contracts. [music] Now, what each side has put on paper. Start with Musk's side. In April, Bloomberg reported that the goal is to begin silicon manufacturing by 2029 and then scale up. Staff had asked equipment makers for prices and delivery times, but no firm orders had been placed. Samsung, which already [music] makes chips for Tesla, reportedly offered extra capacity at its plant in Taylor, Texas, rather than joining directly. And TSMC's chairman CCI said around the same time that a new fab typically takes two to three years to build then another one to two years to ramp up. Next, Intel when it joined its chief executive Lip Bhutan said Terafob represents a step change in how chips will be built. In its own earnings report, Intel calls itself a strategic partner in the project. In the second quarter, Intel's foundry business, the arm that makes chips for other companies, [music] had $5.8 billion in revenue and a $2.1 billion operating loss, down from $3.2 billion a year earlier. Outside customers supplied only a small slice of that revenue. According to the mly fool, Terraab was never part of those numbers. So, Intel is still losing a great deal, but less than before, and a lot is riding on the story. Some commentators go further. Electric wrote that Intel would in effect run the plant. Intel's own statements do not say that. TSMC has not said what it wants. Here is what has been reported. On October 2nd, Tim Culpin's Culpium newsletter reported that TSMC is exploring ways to help run Terraab's planned factories in Texas. In one structure, TSMC owns and operates a new plant with Terraab as its main customer. In another, SpaceX and Terraab hold the majority and TSMC supplies technology and operating knowhow. In September, Reuters had reported, citing unnamed sources, that TSMC was weighing a further investment in Texas on top of the $265 billion it has committed to Arizona. In April, we called Intel a formidable competitor and noted that both Intel and Tesla remain TSMC customers. Musk reportedly replied that Tesla and SpaceX would remain major customers of TSMC, but that TSMC alone could not meet their demand. TSMC declined to comment on the October [music] report. One report said Kulpin himself stressed that nothing is signed. Intel had been the only chipmaker announced as a partner [music] for 6 months. Then on Monday, October 5th, its stock fell more than 4% before the open. One investor news site, Inve reported that Musk later indicated TSMC's [music] chips would likely add to Intel's, not replace them. In the chip business, loyalty is about as thick as a wafer. There is a deeper tension. Terraab is built on vertical integration. One site doing everything for its owners. TSMC's business is the opposite. It makes chips for other companies. Nvidia, Apple, AMD, and Broadcom among them. Business Today reported that TSMC's preferred route could be to fold Terafob into its own Texas expansion with Musk as an anchor customer rather than hand over control of the manufacturing. That is a report, not a position TSMC has stated. Three parties, three positions. [music] Musk has a factory he wants built and chips he says he can't buy. Intel has a technology to show and a very visible partner to show it with. TSMC has experience and capacity and has not said what it wants. In the documents reported so far, none says who runs the room. [music] What holds up? Start with what is confirmed. Intel joined the project in April. As of August, Texas has announced the first phases investment, though that figure has already moved once. Musk has confirmed that discussions with [music] TSMC are taking place. No deal with TSMC has been announced. Then the goals, one terowatt of computing a year. The plan's wafer targets from a pilot line toward a million wafer starts a month. and according to Bloomberg, the start of silicon manufacturing by 2029. Each is a plan stated by the people building it. Then what is next? In October, TSMC reports its quarterly results on the 15th. And according to ticker, Intel's design kit for 14A, the tools chip designers use to build on that process is due. What does each side stand to lose? SpaceX's filing ties Terraab particularly to computing in orbit. If terrafab fails, it says it may not have other sources of enough chips for that. For Intel, it's the showcase. Terraab is its most visible outside partner, and a shared one is a smaller prize. For TSMC, [music] the reported options differ. In one, it builds a new plant in Texas [music] on top of what it has committed in Arizona. In the other, it mostly supplies [music] knowhow. The 2% was Musk's number for the rest of the world's supply. The factory [music] meant to change that is not planned to begin making silicon until 2029 by Bloomberg's account. Bloomberg also noted in April that Musk's companies had never made chips, not once. Weigh the three sides again. Musk can build, but hasn't built this. Intel can make chips, but hasn't said what it's making here. TSMC can do both and has not said what it wants. For months, the question was whether companies that had never made a chip could build what Musk has called the largest chip factory in history. Suppose they can. The question that remains is the one this video began with. If it is built, who gets to run the room? [music]