TSMC Wants In on Elon's Terafab. Intel 14A Stays, and Chip Tool Makers Win Either Way
TSMC is now courting the Terafab plant Elon said it couldn't supply, Intel 14A keeps its seat, and the fab tool makers book the orders whichever foundry signs.
Elon Musk calls it "just discussions." Whichever foundry ends up running Terafab, the machines that fill it come from the same short list of tool makers.
Elon Musk confirmed on Saturday that TSMC is talking to SpaceX about Terafab, the Texas chip plant SpaceX and Tesla announced in March. His reply on X was "Just discussions, but something may come of it." He posted it after Culpium (Tim Culpan's Taiwan chip newsletter) reported that TSMC is exploring ways to work with Terafab. TSMC hasn't commented. Its US-listed shares, $TSM, still closed at a record $485.80 on Monday. $INTC, whose Intel 14A process is the only one Terafab has named so far, slipped about 2%.
Back in April, Elon said Terafab existed because "TSMC just can't make the staggeringly large number of chips needed." Now the world's biggest foundry is asking how it can get involved. SpaceX has become a much bigger buyer since it raised $75 billion in its June IPO, and I think TSMC noticed. If TSMC comes in, Intel keeps its spot and $SPCX gets two ways to fill a plant aiming at 1 million wafer starts a month (a wafer start is one fresh silicon disc fed into the line). Both routes buy their machines from the same handful of tool makers.
What a TSMC role at Terafab would look like
Culpium sketched two versions. In the first, TSMC owns and runs a new Texas plant while SpaceX invests or commits to buying the chips. In the second, SpaceX keeps the majority stake and TSMC brings its process technology and the engineers who know how to run it. The second is closer to the plan Elon laid out in March, where SpaceX and Tesla own the factory and decide what it builds.
TSMC has the money for either one. In July it raised its 2026 capex guidance to $60-64 billion. It also added another $100 billion for Arizona. So TSMC is building more US fabs either way, and a Terafab fab would open with SpaceX as its customer (the same customer who said in April that TSMC couldn't keep up). The full-scale site is the former Gibbons Creek power station in Grimes County, where Texas announced a first phase of more than $16.8 billion in August. Construction is set to start December 1 and run through the end of 2028.
Intel 14A keeps its seat at Terafab
Elon said on Tesla's April 22 earnings call that Terafab plans to use Intel's 14A process. 14A is Intel's next manufacturing generation, and Intel has confirmed it'll use ASML's High-NA EUV machines to make it (EUV prints the circuit pattern with extreme ultraviolet light, and High-NA is the newer version with a wider lens that prints finer lines). Intel badly wanted that customer. Its own quarterly filing raised the prospect of pausing or discontinuing 14A without a significant outside buyer. CEO Lip-Bu Tan has since talked about working with Elon to "refactor silicon process technology."
You can see why Intel shares dipped. SpaceX's S-1 (its IPO registration filing) says "neither Tesla nor Intel are obligated to remain a part of the project," which reads to me like a buyer keeping its options open. This week SpaceX used one of them. But the two foundries' calendars line up. Intel 14A targets risk production in 2027, with volume a year later (risk production means early runs for customers who'll accept low yields). TSMC's A14 is on the same schedule. A plant chasing 1 million wafers a month has room for two process recipes, and I think having both makes SpaceX's timeline safer. Elon would get to compare two foundries' yields (the share of chips on each wafer that work) side by side in one plant, and I can't think of another US fab set up that way.
The fab tool makers that book the Terafab orders
Back in April, SpaceX and Tesla teams asked Applied Materials, Tokyo Electron (Japan's largest chip-tool maker), Lam Research and Samsung for prices and delivery times on photomasks, cleaning gear, testers and the etch and deposition tools that build each chip up layer by layer. In June Elon joined ASML CEO Christophe Fouquet remotely for a fireside chat at ASML's employee conference, where ASML said many companies, including ASML, will collaborate on the project. Each High-NA machine runs about $400 million. That's roughly twice the price of a standard EUV tool, so even a few of them add up fast on ASML's order book.
Whoever wins the process role, the tool makers see it first, as orders in their backlog long before a single Terafab chip ships. $ASML is the only company that sells EUV machines, so its quarterly net bookings are the first place a Terafab order would show. $AMAT, $LRCX and Tokyo Electron were already pricing Terafab gear in April. $KLAC hasn't been named in any Terafab report, but I can't think of a leading-edge fab that ramps without its inspection tools checking every layer. We covered the power and cooling side of the same build in our earlier Terafab suppliers piece.
TSMC reports third-quarter results on October 15, and TrendForce lists the Terafab talks among the themes to watch on that call. I'm watching ASML's bookings line more than either foundry's answer. If TSMC joins, Terafab has better odds of hitting its dates, while Intel staying gives SpaceX a second source. So I see SpaceX holding the stronger position in both negotiations, with ASML, Applied and Lam booking the orders whichever foundry signs. Nobody's disclosed a tool order yet.
Elon wants 1 million wafers a month out of Texas. Whichever foundry signs, ASML's order book is where that number shows up first.