SpaceX is considering spending up to $119B on a semiconductor manufacturing complex in Grimes County, Texas, according to a proposal posted on the county's website. The initial phase alone is budgeted at $55B, making the project — internally dubbed Terafab — one of the largest single private capital commitments ever floated for a chip fab. The filing describes a "multi-phase, next-generation, vertically integrated semiconductor manufacturing and advanced computing fabrication facility" intended to supply Elon Musk's expanding constellation of AI and robotics ventures.
The chips are meant for xAI's Grok training and inference clusters, Tesla's autonomous vehicles and humanoid robots, SpaceX satellites, and a proposed orbital data center. Musk has set a long-term output target of 1 terawatt of chip-driven compute per year, a figure that would dwarf the combined AI-accelerator output of every existing foundry. Intel has been pulled into the project as a manufacturing partner, with Tesla also contributing resources.
"We either build the Terafab or we don't have the chips, and we need the chips, so we build the Terafab," Musk wrote, framing the project as a forced response to what he sees as insufficient supply from existing foundries. The argument echoes a recurring Musk theme: that vertical integration is cheaper than waiting in line behind hyperscaler customers at TSMC and Samsung.
Key facts
- 01SpaceX estimates total spending of up to $119B on the Terafab semiconductor project, with $55B earmarked for the initial phase.
- 02A Grimes County, Texas filing describes Terafab as a multi-phase, vertically integrated chip manufacturing and advanced computing facility.
- 03Musk has set a long-term target of 1 terawatt of chip output per year to supply xAI, Tesla, and SpaceX.
- 04The combined xAI-SpaceX entity is valued at $1.25T and is expected to go public in June.
- 05Intel has been roped into the effort, with chips earmarked for AI servers, satellites, an orbital data center, and Tesla robots.
The filing arrives at a delicate moment. Musk has merged xAI into SpaceX, and the combined entity is reportedly valued at $1.25T with an IPO expected in June. A $119B capex commitment, even staged, would be a defining line item in that prospectus — and a tangible justification for the merger, which Musk has framed around the compute demands of training Grok and operating data centers in space.
“We either build the Terafab or we don't have the chips, and we need the chips, so we build the Terafab.”— Jaeden Schafer
Grimes County may not end up being the site. Musk wrote on Tuesday that the Texas county is only one of several locations under consideration, a hedge that gives SpaceX leverage on incentives and permitting. Filings of this scale typically anchor negotiations with state and local officials over tax abatements, water rights, and grid interconnects — all of which become acute at terawatt-scale power draw.
For Intel, the partnership is the more consequential bet. Intel's foundry business has struggled to land marquee external customers, and a SpaceX-Tesla-xAI anchor tenant at Terafab scale would reset the narrative. The structure of the deal — whether Intel operates the fab, licenses process technology, or co-invests — has not been disclosed in the Grimes County filing.
The project also reframes the AI-infrastructure arms race. OpenAI has leaned on Microsoft and Oracle for compute, and is backing Cerebras, which is on track for an IPO. Anthropic depends on Amazon and Google. Musk's pitch is that owning the fab — not just renting GPUs — is the only durable answer to chip scarcity. Samsung, which just crossed a $1T valuation on the back of AI memory demand, is the cleanest counter-example of how lucrative the supply side has become.
The skeptic's case writes itself. Greenfield fabs routinely run years late and tens of billions over budget; Intel's own Ohio project is the cautionary tale. A 1-terawatt output target has no precedent and no clear timeline, and Musk's tweet acknowledging multiple candidate sites suggests the Grimes County filing is closer to a feasibility envelope than a binding commitment. The capital stack to fund $119B — even staged across a decade — is not yet visible in any public xAI or SpaceX disclosure.
There is also the question of whether public-market investors will underwrite the spend. The combined xAI-SpaceX IPO planned for June will ask buyers to fund rockets, satellites, frontier AI training, an orbital data center, and now a fab the size of a small country's industrial base. Each leg has its own capex curve and its own probability of slipping.
Terafab, if it gets built at anything close to the filed numbers, would mark the moment AI infrastructure stopped being a software-and-GPU story and became a heavy-industry one. The constraint on frontier models has shifted from data to compute to power to silicon supply, and Musk is betting that whoever owns the bottom of that stack ends up owning the economics of the top. The risk is that the fab arrives years after the model race has already been decided — and that $119B buys a monument to a problem someone else solved with cheaper means.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
