Terafab is the massive, vertically integrated semiconductor fabrication complex announced by Elon Musk on March 21, 2026, jointly developed by Tesla, SpaceX (including xAI), and Intel. It represents the critical manufacturing foundation to produce the specialized AI chips needed for both terrestrial applications (Optimus robots, vehicles) and, most importantly, orbital AI infrastructure.
Here’s the size, scale, and forward plan:
• Unprecedented production target: Terafab aims to manufacture over 1 terawatt (1 TW = 1,000 GW) of AI compute capacity per year — logic chips, high-bandwidth memory (HBM), and advanced packaging all under one roof. This is roughly 50x current global advanced chip output and double the entire U.S. electricity consumption scale.
• Physical footprint: A sprawling complex spanning thousands of acres in Texas (multiple candidate sites near Austin, including Grimes County considerations), far larger than Giga Texas combined. It includes a dedicated research/advanced technology fab for ultra-rapid chip iteration (new designs potentially every day) plus two high-volume production fabs.
• Power hunger at scale: The full facility will require over 10 GW of dedicated power — an entire power-plant-scale demand on its own.
• Investment scale: Initial phases in the tens of billions, with total project costs potentially reaching ~$119 billion. It is designed as a closed, vertically integrated system optimized exclusively for Musk’s ecosystem needs (no external sales).
• ~80/20 split favoring space: Roughly 80% of output dedicated to space-optimized AI chips (D3 series — radiation-hardened, designed to run hotter to slash radiator mass/weight), with ~20% for Earth-based uses like Optimus and vehicles.
The orbital AI compute scaling plan (enabled directly by Terafab’s chip output Starship launch capacity):
• Why space wins: Earth faces hard limits on power generation, grid infrastructure, land, and cooling. Space offers unlimited solar power and perfect radiative cooling in vacuum. AI compute there can run at massive scale without terrestrial constraints.
• Specialized hardware: Custom D3 chips optimized for orbit (higher operating temperatures dramatically reduce the size and mass of giant heat radiators — a make-or-break factor for launch economics).
• Launch vehicle enabler: Starship’s enormous payload capacity and reusability make it possible to loft the required millions of tons of solar arrays compute hardware per year.
• Prototype foundation: 100 kW “AI Mini Sat” concepts already shown (solar arrays radiators). Future versions scale to megawatt-class modules.
• Ramp plan as described: Terafab’s output will support shipping 1 GW of AI compute capacity into space next year (2027). From there, the plan scales by a factor of 10 every other year — creating explosive exponential growth in orbital intelligence infrastructure (1 GW → 10 GW → 100 GW → 1 TW within a decade).
• Ultimate vision: Launch millions of tons of solar capture hardware annually to power terawatts of space-based AI. Synergies include billions of Optimus robots building everything on the ground (and potentially assembling in orbit) while Starship handles transport. This is framed as the missing piece for humanity becoming a true multi-planetary (and eventually galactic) civilization.
Terafab is not just another chip factory — it is the industrial engine designed to flood space with AI compute at a scale that makes today’s Earth-bound data centers look like pocket calculators. The combination of Terafab’s 1 TW/year chip production Starship’s launch cadence purpose-built orbital hardware creates a credible path to exponential scaling of AI capability in orbit, starting with 1 GW shipped next year and then multiplying by 10x every two years. This is the infrastructure layer for AI that can truly scale without Earth’s energy ceiling.