How Elon Musk Is Building a $119B Factory That Might Not Even Be Real

Kind: captions Language: en It doesn't exist yet. $119 billion, 100 million square feet, 10 times the size of TSMC's largest chip factories— the plants that already produce almost every modern processor on Earth. Not a single chip manufactured, not a single factory launched. Here is Musk's real experience in this field. I read the government report on this project twice. It all started with an announcement on stage in Austin, and now the project is under the watchful eye of his own company's lawyers. Almost 900 residents near the construction site signed a petition before the first wall was erected . And here's the strangest thing. The same company building this plant just wrote to US regulators that all of this is not even a promise. So why believe? A tweet or an official report? Terafab. 119 billion dollars. This is the upper limit of cost that Elon Musk has set for a single building. 100 million square feet. 10 times more than the largest chip factories TSMC—the companies that already produce almost every modern processor on Earth. Musk announced more than just a factory. He announced the largest and most expensive building ever built by humans. And 3 months later, his own company submitted a report to regulators that said something completely different. I read this document myself. Page after page, lawyers refuted what Musk said from the stage in Austin. The enemy here is not sand, not sea water, or faults in the earth's crust. This is a learning curve that took Taiwan and Korea three decades to overcome. And a company with zero experience in chip manufacturing is trying to go this route in less than 3 years. 10 minutes from her home, Suzanne Abney learned what was brewing on her land . When she bought the property, she was not told anything. Almost 900 of her neighbors signed a petition before the first foundation was poured. Not against jobs, but against silence. That's why it even exists. Tesla, SpaceX, and the newly formed xAI argue that the world's chip factories can only supply a fraction of what their robots, autopilots, and orbital data centers will need. So instead of waiting in line for Apple or Nvidia, Musk decided to build his own line—from raw silicon to finished chip under one roof on a scale that no factory in the world had yet dared to implement. The foundations are already being laid in Austin . A full-scale project is just a date on a government report, but how do you build the largest plant on Earth when the company building it just told the US government in writing that it was n't even a promise? Stand in a real chip manufacturing “clean room” for 5 minutes and you will understand why almost no one takes on it. One speck of dust can ruin a plate worth thousands of dollars. It is an invisible enemy, not sand or seawater. A manufacturing process that one analyst called the furthest point humanity has advanced on the technology tree. That's the scale of the problem. It took TSMC over 30 years to reach today's production output rates. Samsung, a company that has been making chips for decades, still only has about 55% yield on its newest 2- nanometer process. TSMC reaches over 80%. This gap in the process, which both companies have been using for years, is the difference between profit and loss. Musk's companies have not produced a single chip. In general . And here's the craziest part. His answer was not to hire experts to bypass the learning curve, but to redesign the problem itself. “Divide the factory in half,” he said. “One factory, one chip design for each. No branches, no cross- contamination between production lines. " Only wafers moving in a straight line from raw silicon to finished chip." He called it linear flow. I've talked to engineers who have spent their entire lives in this field. Their reaction to this idea was divided. Nvidia's own CEO, a man who has never publicly opposed Musk, clearly stated: " This kind of process expertise takes years of work." No company develops it overnight." Now look what they decided to do with it . March 21, 2026. A power plant in Austin has been decommissioned. Musk stood before her and gave the project a name—“Terrified Planet.” Musk continued: “Aerafab, without outside experience, it will fall apart before it even starts. So 17 days later, Intel joined in, providing 14A, one of the most advanced manufacturing processes being developed on Earth. Not a deal with a supplier, but a production partner inside the walls. This is the foundation that no one sees in visualizations. A 1.4- nanometer process that even Intel hasn't yet perfected . To run it, they needed someone who had done it before. Gary Gang, 17 years at Intel, an engineer who led the launch of Intel's own 18A fab in Arizona, was brought in to lead Terafactoring. As a former Intel employee, Gang knew Rayfab herself. That's why it was important. Skip the capacity-building stage in such a complex process, and the factory won't just produce chips more slowly. It will not produce anything. The following month, work began on the first part that anyone could actually see—a prototype factory at Tesla's Giga Texas campus. This is not the entire 100 million square foot complex, but a testing ground to test the process before putting everything on the line. Stand by that fence today, and it will look like any other construction site. Steel, concrete mixers— nothing yet points to the largest building on Earth. Two factories, one manufacturing partner with no proven history on this node, one engineer whose career led to this point . This is the foundation upon which Terafab was actually built. Now look at what they said the finished object should have looked like. This is what they say they will build. Two chips: one for Earth, one for space. The ground chip, which Tesla calls AI 5, consumes 150 watts. A similar Nvidia chip, the H100, requires 700 watts for the same performance. That's a 4.7x difference in efficiency, if the numbers are correct. Look here. AI 6, the next chip in the line, is designed to double the performance of AI 5 in the same physical size. It is manufactured at Samsung's Taylor, Texas, 2nm manufacturing facility, alongside a separate version of AI 6.5 at TSMC in Arizona. Two foundries, two versions of the same chip, so that a delay on one doesn't stop the other. This dual source of supply is standard in the industry. What is not standard is when a company does this for a line of chips that do not yet have a factory. The second chip also has a name. D3. Radiation-proof, designed for orbit, it will power the AI ​​satellites that SpaceX wants to use for its space data centers. This is a figure that goes beyond any scale. Musk's goal: 1 million wafer launches per month by 2030. TSMC, the company that already makes nearly every advanced chip on the planet, is currently producing about 1.42 million wafers per month across its global network. One new company aims to achieve approximately 70% of this figure on its own at a single facility. I read it three times because I didn't believe it the first time. 100 million square feet to accommodate it all. 10 times the area of TSMC's largest factories in Arizona. Factories that cost TSMC $165 billion to build. Terafab's claimed area dwarfs all six combined. No one has ever run a semiconductor company of this scale with such a small manufacturing history. Neither TSMC in its early years, nor Samsung, nor Intel. But there was one problem that no one could get around, no matter how good the chip roadmap looked on paper. Then the documents appeared. On May 20, 2026, two months after the announcement in Austin, SpaceX filed an IPO prospectus with US federal regulators. 308 pages where, by law, investors are required to be told the truth, not marketing promises. Inside, Terafab—the same project that Musk called the largest building on Earth— was described only as a general concept. No binding agreements between Tesla and SpaceX. No final distribution of intellectual property. No obligation for any company to continue participating. I read this paragraph twice. This was not the blogger's opinion. These were SpaceX's own lawyers , under legal obligation to refute what their CEO had just told the world. The skepticism didn't end there. Nvidia CEO said Musk underestimates a process that takes years to master. The Electrek editor called the entire statement, in his own words, an act of desperation. One semiconductor market analyst put it in numbers. There's an 80% chance, he said, that Terafab will never appear. Here was the difference that mattered most. Musk needed a project of this scale to justify the trillion- dollar valuation. SpaceX's own lawyers needed to protect new investors from a promise that was not backed by a contract. Both documents came from the same company. Both could not be entirely true. Nothing was decided. The discrepancy between the statement and the official report has become official. Recorded in the protocol. Forever. At the scene of the incident. Something has moved. On August 5, 2026, SpaceX representatives finally met face-to-face with the residents of Grimes County for the first time. They made specific commitments. No increase in local electricity rates, no depletion of groundwater, no draining of the Navasota River. Their own gas-fired power plants will supply the site, not the general grid. One resident told me the hall was the most packed she had seen at a county meeting in years. This was a meeting that people had been waiting for for 5 months. The next day, commissioners approved the deal and the first phase, worth $16.8 billion, as well as a full tax exemption that same week . Nearly 900 residents filed a petition demanding 12 specific guarantees in writing. Here's what remains unresolved. As of this month, air emissions and water use permits that would make these verbal promises legally binding have still not been submitted. A promise on stage. A promise behind the podium. There is still no permit on paper. Then it happened. September 11, 2026 . A drone soared over Giga Texas' north campus. No announcements, no press releases, just a private individual, an aviation photographer named Joe Tegmeyer, making a routine flyby of an object he had been following for months. What did the video show? Poured concrete. A real foundation already taking shape at the prototype plant less than 6 months after the announcement in Austin. I compared these shots with construction photos of other semiconductor projects. Most plants don't show visible foundation work this early. This one showed. Not the entire million square foot complex. This site in Grimes County is still just a date in the report. December 1. But on Giga Texas, right now on site , the concrete has already been poured. No lawsuit can argue with the foundation. The documents refer to general frameworks. The earth says that something is actually being built. Both statements are correct at the same time. This is Terafab. 119 billion dollars. 100 million square feet. A building that doesn't yet fully exist, and a report that acknowledges that it may never exist. Nearly 900 Grimes County residents didn't sign that petition because they believed the hype. They signed it because they learned not to believe promises until they are on paper. Suzanne Abney still lives 10 minutes from that fence. Her land did not move. Everything around her is like that. Maybe this is the real story. It's not about whether Elon Musk will build the largest building on Earth, it's about whether anyone will distinguish between a company that pours concrete and a company that pours confidence until the plates finally start coming out the other end. The foundation is now real. The promise—still not. If Terafab is ever completed, it will change the world's chip sourcing. If not, it will become the most expensive filing cabinet in Texas. In any case, keep an eye on this place. Literally.