Tesla and SpaceX confirmed Grimes County, Texas, outside Houston, as the site for Terafab, committing $16.8 billion to a first phase with more than 100 million square feet of manufacturing space planned and at least 3,000 jobs, per TechCrunch. Civil work starts almost immediately, first chips are targeted for late 2027, and volume production is not expected before 2028, per Electrek. The campus is planned as two fabs, each dedicated to one chip family: the AI5 and AI6 architectures behind Tesla's Full Self-Driving stack, Optimus, and Cybercab, and a radiation-hardened D3 part for SpaceX's orbital data centers.
The detail most coverage buries is who is actually doing the manufacturing. Intel Foundry is the process partner, contributing its 18A node, and analysts have read the structure as closer to an Intel fab expansion with Tesla, SpaceX, and xAI as anchor customers than a Musk-run semiconductor operation, per Electrek's earlier reporting. That reframes the announcement. Musk is not learning to run a leading-edge fab, which is the single hardest operational problem in technology. He is buying guaranteed capacity on someone else's process and putting his name and capital on the building.
For Intel this is the more consequential half of the news. 18A uses gate-all-around transistors and backside power delivery, and Intel Foundry has spent two years hunting for the external anchor customers that would justify it against TSMC. Three Musk companies committing to the node, at this scale, is the validation the foundry business needed, per EE Times. It also means the schedule now depends on 18A yields, which were reported near 65% at Intel's Arizona plant in April. Yield is what separates a fab announcement from shipped silicon.
Two other specifics are worth holding onto. Terafab is planned to include packaging and testing alongside wafer fabrication, and packaging is the actual bottleneck in AI hardware right now, the constraint behind TSMC's sold-out CoWoS lines and Nvidia's $1.5 billion prepayment to Amkor. And the power plan is natural gas plus large battery arrays rather than Tesla solar, per TechCrunch, which tells you the priority is firm power on a 2027 schedule over brand consistency.
The context is SpaceX's own quarter. It spent $15.83 billion on AI infrastructure in three months against $7.8 billion of revenue, and the market took 24% off the stock. Terafab is the vertical-integration answer to that math, an attempt to own the supply of the chips whose scarcity is inflating those bills. The cost is another multi-year capital commitment layered onto a company already burning cash on compute.
Bottom Line
Read this as Intel's largest foundry win as much as Musk's fab. The schedule rests on 18A yields and on packaging capacity, so watch Intel's yield disclosures rather than the groundbreaking photos.