SpaceX filed regulatory paperwork last week for a $55 billion semiconductor fabrication plant in rural Texas, bringing the company's total disclosed chipmaking infrastructure spend to $119 billion when combined with the existing Bastrop packaging facility. The figure is five times larger than the $20 billion estimate Elon Musk cited when he announced Terafab in March. The filing names the project simply as "Terafab" and lists completion targets in phases through 2029.
The Bastrop packaging site went operational in late 2023 and currently processes custom ASICs for Starlink's third-generation satellites. The new fab filing shows SpaceX intends to manufacture those chips domestically rather than rely on TSMC or Samsung contract runs. The $55 billion fabrication plant would sit roughly forty miles from Bastrop, with early-stage site preparation beginning in Q2 2025 and first wafer production scheduled for Q4 2027. SpaceX has not commented publicly on the filing. The paperwork was required under Texas environmental impact rules for projects exceeding $10 billion in capital expenditure.
The jump from $20 billion to $119 billion suggests either significant scope creep or deliberate misdirection in the March announcement. Starlink currently operates over 5,400 satellites and plans to reach 42,000 by 2030 under FCC filings. Each third-generation satellite uses three custom ASICs for phased-array antennas, on-board routing, and laser inter-satellite links. At that scale, captive fabrication starts to make economic sense if wafer costs drop below $4,000 per unit, which is achievable at the 5-nanometer node SpaceX appears to be targeting. TSMC charges closer to $16,000 per wafer for similar processes when orders fall below 50,000 wafers per quarter.
The strategic read is vertical integration at a pace the semiconductor industry has not seen outside of Intel's 1990s build-out or Samsung's 2010s push. SpaceX would control chip design, fabrication, packaging, and deployment — the entire stack from sand to orbit. That eliminates supply-chain risk, which has already delayed Starlink launches twice in 2024 due to ASIC shortages. It also creates a chipmaking competitor with $119 billion in sunk capital and no need to sell to outside customers, which changes margin assumptions for anyone bidding on future Starlink contracts. The Bastrop facility already poached twelve senior engineers from Intel's Arizona fabs between June and August, according to LinkedIn employment records.
Operators should watch Texas state permitting through May, when environmental review closes and construction bonds get finalized. If SpaceX moves dirt before June, the 2027 production target is credible. If permitting drags past July, first wafers slip to 2028 and the $119 billion figure becomes a negotiating position rather than a committed budget. TSMC's Arizona fab faced sixteen months of permitting delays and came in 38% over budget. Starlink's next-generation satellite architecture, expected to be disclosed in Q3 2025, will clarify whether this fab is defensive or expansionary — whether SpaceX is just insourcing existing demand or preparing to flood low-earth orbit at a pace that requires twice the chip supply anyone currently models.
The filing also lists "AI accelerator production" as a secondary use case, buried in the environmental impact section under projected energy consumption. That line did not appear in any prior Terafab discussion.