systems

Musk has never built a wafer fab, but he wants to burn $119B on one anyway

Initial phases of SpaceX's Terafab project in rural Texas are expected to cost about 1.25 Twitters

Elon Musk's SpaceX hopes to plow as much as $119 billion into the southeast Texas countryside to build a massive semiconductor fab to produce chips for orbital AI datacenters. We won't hold our breath.

The harebrained scheme is part of Elon Musk’s recently announced Terafab project, which seeks to boost global semiconductor production by 50x. After all, how else is he supposed to lob a terawatt of compute a year into orbit?

In a recent public filing, SpaceX described the project as a “multi-phase, next-generation, vertically integrated semiconductor manufacturing and advanced computing fabrication facility.” 

The facility would be located roughly 80 miles northeast of Houston near the Gibbons Creek Reservoir. 

The first phase of this project is expected to cost around $55 billion, or about 1.25 Twitters. For reference, Intel’s leading-edge fab expansion in Chandler, Arizona cost roughly $30 billion, but that facility pales in comparison to Musk’s ambitions.

In addition to silicon used for compute, Musk claims Terafab will have all the equipment necessary to produce chips of any kind, and that includes memory.

“In a single building, we can create a lithography mask, make the chip, test the chip, make another mask, and have an incredibly fast recursive loop for improving the chip design,” Musk boasted in a March presentation.

That’s an ambitious plan from a man who has never run a fab before.

SpaceX, which now includes Musk’s AI startup xAI and by extension the shriveled husk of the once great social network Twitter, knows an awful lot about building launch vehicles, satellites, bit barns, and web-scale applications. But last time we checked, the companies have zero semiconductor manufacturing experience.

This knowledge gap doesn’t seem to have Musk all that worried, seeing as last month American foundry partner Intel just signed on to help out with the project.

During Tesla’s most recent earnings call, the CEO revealed his Terafab manufacturing venture would produce chips based on Intel’s yet unfinished 14A process node.

These chips will include homegrown AI accelerators - an area where Tesla, also part of the Terafab project, does have some experience, unlike chip manufacturing.

Tesla has developed several generations of custom silicon to power its electric vehicles, as well as its fully custom Dojo supercomputing platform. So at least that bit is plausible. 

Of course, demand for these chips is predicated entirely on SpaceX’s Starship bringing down the cost to orbit to a level where orbital datacenters are economically viable, which still hasn’t happened yet.

But seeing as it takes three to five years to bring a new fab online, SpaceX still has time to make its first truly reusable rocket… well, reusable.

Musk has a history of making big promises and then not delivering on them. You might remember him promising to cut $2 trillion and then $1 trillion from government spending via DOGE, but only managing to cut about $150 billion or 2.2 percent. He has also predicted that SpaceX would send a rocket to Mars by 2024 and that Tesla would have a million robotaxis on the road by 2020.

In the meantime, the Grimes County Court of Commissioners will consider whether to grant Musk’s Terafab project a property tax abatement during a meeting at 9 am on June 3. ®