Tesla and SpaceX picked a site for Terafab and committed $16.8 billion to start
A retired coal plant in Grimes County becomes a chip fab aimed at a terawatt of compute a year, for Optimus robots and orbital data centers.
The figure has been $25B, then $119B, now $16.8B. Reading the phases matters more than the headline.
SpaceX and Tesla announced jointly on August 6, 2026 that Terafab, the chip fabrication plant they are developing together, will be built in Grimes County, Texas, about an hour northwest of Houston, on the site of a coal plant that closed in 2018. The initial combined investment is $16.8 billion, the per-company split was not disclosed, and the stated plan is a vertically integrated complex of more than 100 million square feet making, packaging, and testing advanced logic and memory chips, with at least 3,000 local jobs.
Read the number carefully
The $16.8 billion is a first phase, not a total, and the figures attached to this project have moved constantly. It was reported at roughly $25 billion in March. SpaceX's May S-1 filing described Terafab as a general framework with a proposed $55 billion initial investment rising to $119 billion across phases, and characterised the arrangement as non-binding with the intellectual property split unresolved. That filing is the most useful document here precisely because it is the least promotional one. No construction or production timeline was disclosed in this announcement, and Intel's previously reported manufacturing role went unmentioned.
What it is for
The stated target is over 1 terawatt of compute per year, a company claim with no published methodology behind it, aimed at Tesla's Optimus robots and Cybercabs and at high-power chips for SpaceX's planned space-based data centers. Musk described it as the largest and most valuable building on Earth by far. Water for the site comes from the Gibbons Creek Reservoir rather than local groundwater, which is a more consequential detail than it sounds given how much of the current data center backlash in Texas is about resource contention.
Why a build studio cares
The interesting part is not the building, it is the decision. Two companies with unusually good access to the chip market have concluded that buying on that market will not meet their own compute demand, and have chosen to vertically integrate into fabrication instead, which is the single most capital-intensive and slowest-moving thing you can choose to do. Tesla already holds a $16.5 billion foundry contract with Samsung, so this is not for want of a supplier. It is a statement that they expect supply to stay tight for longer than a contract can cover.
Next step: read TechCrunch's report and Electrek's more sceptical read. If you want a build plan that does not depend on infrastructure that has not been built yet, write to us at hello@gattyworks.com.