Musk’s Terafab: SpaceX Plans $119B Texas Chip Megafab for AI
3 min readElon Musk’s SpaceX filed paperwork this week outlining a Texas chip plant that could grow into the largest semiconductor build in U.S. history. The first phase is pegged at $55 billion. The full buildout could reach $119 billion. The target customer is not consumers or phones. It is the AI compute appetite of Musk’s own companies, and increasingly, his rivals.
What was filed
According to TechCrunch and CNBC, SpaceX disclosed plans for a vertically integrated semiconductor and advanced computing facility branded “Terafab.” Grimes County, Texas is the leading site, with $55 billion locked in for the first phase and a multi-phase ramp that could carry the project to $119 billion. The campus will combine logic, memory, and advanced packaging on one site.
Intel is on board as a manufacturing partner, contributing its 18A process node. That makes Terafab the most ambitious U.S. fab project announced this decade, dwarfing recent Arizona and Ohio buildouts in scale and intent.
Why a rocket company is building chips
Musk has spent the past two years scrambling to feed three different AI workloads. xAI needs training and inference capacity for Grok. Tesla needs custom silicon for the Optimus humanoid program and the Robotaxi fleet. SpaceX has its own pipeline, including a radiation-hardened processor reportedly called the D3 that is meant to run AI workloads in orbit on Starlink and future data center satellites.
Buying merchant GPUs from Nvidia at frontier-lab prices does not scale to that footprint. Owning the fab does. By tying Terafab to Intel’s 18A process and pulling logic, memory, and packaging under one roof, Musk is trying to do for AI silicon what SpaceX did for launch: collapse the supply chain and the unit cost at the same time.
The Anthropic angle
The most striking detail is who else gets to use the output. xAI has already cut a deal letting Anthropic raise its compute usage limits on xAI infrastructure, repositioning xAI as much as a “neocloud” landlord as a frontier model lab. If Terafab delivers, the same logic extends to silicon: Musk would be selling chips, capacity, or both to the very companies competing with Grok and Optimus.
That is a notable shift in how the AI industry talks about itself. The race used to be framed as model versus model. It is now openly a race for the underlying compute, and Musk is positioning to be the toll-taker rather than just a participant.
What to watch
Two things will decide whether Terafab is real or theater. First, the timeline: Morgan Stanley estimates that even an aggressive build would not produce first chip output until mid-2028. Second, the financing path. The $55B first phase is large enough that SpaceX will likely need outside capital, federal CHIPS Act incentives, or both. Texas approvals, water and power agreements, and Intel’s 18A yield curve will all be early signals.
If it lands, Terafab will reshape the AI supply chain the same way Anthropic’s Google and Broadcom deals reshaped the demand side. If it slips, it joins a long list of headline-grabbing fab announcements that quietly shrank.