A semiconductor megafactory spanning 100 million square feet, with projected long-term investment of up to $119 billion, is designed to mass-produce AI chips no bigger than a fingernail. This is not an internet exaggeration, but Terafab, the giant wafer fab jointly developed by Tesla and SpaceX. Elon Musk warns without in-house chip capacity, Tesla, SpaceX and xAI will face severe chip shortages. Nevertheless, this largest semiconductor investment in human history faces widespread industry scepticism, with experts describing its difficulty as “harder than landing on Mars.”
Project Overview: World’s Largest Single Building in Texas
Terafab is located in Grimes County, Texas, and broke ground in August 2026. Built in phases, the initial investment stands at $16.8 billion, with total long-term capital expenditure capped at $119 billion. Its manufacturing floor area reaches 100 million square feet (around 9.3 million square metres), equivalent to 50 Pentagon buildings. Upon completion, it will become the world’s largest single building by floor area.
Unlike conventional fabs, Terafab is conceived as a vertically integrated semiconductor complex. It consolidates chip design, wafer fabrication, DRAM memory production, advanced lithography, packaging and testing on one site, aiming for full end-to-end independence and reduced reliance on external foundries.
The plant plans to adopt Intel 14A (approx. 2nm) advanced process technology to produce two core families of AI chips:
1. Ground AI5 chips: Fingernail-sized processors for Tesla FSD autonomous driving, Cybercab robotaxis and Optimus humanoid robots.
2. Radiation-hardened space-grade D3 chips: Built for Starlink satellites to withstand cosmic radiation and extreme temperature fluctuations.
Strategic Rationale: 1 Terawatt Compute Target, 80% Capacity for Space
Musk’s team estimates Tesla, SpaceX and xAI will require 1 terawatt (1TW) of compute power in future. Current global chip capacity can only supply roughly 2% of this demand, which is the core motivation for Musk’s multi-billion-dollar in-house fab.
Under the capacity allocation plan, once Terafab hits full production, around 80% of compute output will serve SpaceX space operations, powering orbital AI data centres and on-board Starlink satellite inference. Only the remaining 20% is reserved for Tesla’s terrestrial business, including self-driving, humanoid robots and xAI large language models.
The facility targets a maximum monthly wafer output of one million wafers, producing hundreds of billions of tiny fingernail-sized AI chips per year. Viral online claims that Musk “spends billions for one single chip” are clickbait simplification. The true goal is mass high-volume manufacturing, to sustain the long-term roadmap of robots, robotaxis and space-based AI networks.
Industry Divide: Grand Vision Amid Formidable Hurdles
Terafab has sparked heated debate since its announcement, with both supporters and sceptics in the semiconductor sector.
✅ Optimistic view:
If successfully completed, Terafab will secure internal computing supply, drastically cut the cost of custom silicon and enable Musk’s ecosystem to build a dual ground-and-space AI computing backbone, reshaping the global AI chip supply landscape. Intel’s CEO describes Terafab as a revolutionary attempt to redefine chip manufacturing.
⚠️ Industry concerns:
1. Equipment constraints: Advanced nodes rely on ASML EUV lithography systems with long lead times and tight supply.
2. Yield challenges: Yield ramp for 2nm-class advanced processes often takes years. Jensen Huang of NVIDIA has publicly commented Terafab will struggle to match TSMC yield levels. Musk’s “wafer isolation” concept is questioned by specialists, who argue it underestimates strict cleanroom requirements for cutting-edge semiconductor production.
3. Talent & capital pressure: Advanced fabs require tens of thousands of seasoned semiconductor engineers, while global talent is already scarce. Sustained trillion-scale spending may create heavy depreciation burdens if future product demand falls short.
4. Demand assumptions: The whole project hinges on mass adoption of FSD, Optimus robot rollout and scheduled deployment of AI-enabled Starlink satellites. Delayed commercialisation could leave the megafactory with massive idle capacity. Bernstein analysts warn Terafab may be harder to realise than a Mars landing.
Conclusion
Terafab represents the boldest computational gamble in semiconductor history. Enormous plant scale and multi-hundred-billion investment ultimately serve to produce tiny individual AI chips. The outcome of this experiment will shape not only Tesla and SpaceX’s commercial footprint but also the future of AI, autonomous mobility and space computing. Full buildout will take up to 11 years; full capacity remains a long-term vision rather than near-term reality.