SpaceX and Tesla Are Building the "Largest Building on Earth": What Is the $16.8 Billion Terafab Fab Really For?

One-sentence takeaway: Terafab isn't just another Musk-era factory — it crams chip design, fabrication, and packaging under a single roof, targeting more than 1 terawatt of AI compute per year, roughly 50× today's global AI chip output.

In August 2026, SpaceX and Tesla announced a joint venture to build the AI chip factory "Terafab" in Grimes County, Texas, backed by $16.8 billion in initial investment. Elon Musk wrote on social media: "Terafab Texas will be the largest and most valuable building on Earth by far." The complex is expected to exceed 100 million square feet — larger than any semiconductor facility ever attempted. To put that in perspective, it's bigger than Giga Texas, the Pentagon, Apple Park, and the Mall of America combined.

One Factory, Not Just Chips

A traditional fab does one thing: manufacture chips. Terafab brings chip design, mask making, fabrication, memory, advanced packaging, and testing under one roof. Engineers can design a chip, build it, test it, and improve it on-site — compressing a cycle that usually takes months and spans multiple companies and countries.

Those chips will feed three battlefronts of the Musk ecosystem: Optimus robots, Cybercab self-driving vehicles, and space-based data centers. Roughly 25% of output is expected to power Tesla's Optimus robots, Full Self-Driving vehicles, and Cybercabs. The other 75% targets SpaceX's solar-powered AI satellites and orbital data centers — where compute is no longer limited by Earth's power grid.

1 Terawatt a Year: Why Build It Yourself

Musk's rationale is blunt: existing chip suppliers simply can't produce enough AI chips for where Tesla and SpaceX are headed. Terafab's long-term goal is more than 1 terawatt of AI compute per year — roughly 50× today's global AI chip output. At that scale, buying from any external supplier would be impossible.

The consortium isn't just Tesla and SpaceX — xAI and Intel are also involved, suggesting Terafab may support multiple chip architectures and lean on Intel's manufacturing expertise. One detail has drawn particular attention: a ring-shaped accelerator sits at the center of the concept art. Musk replied "FEL FTW" — free electron laser — hinting Terafab may bypass traditional EUV lithography and pursue accelerator-based chipmaking, potentially loosening the industry's dependence on ASML.

Strategy: Compute as Infrastructure, on Par with Launch Pads

When a rocket company starts building a chip fab, the signal is clear: in Musk's blueprint, AI compute is strategic infrastructure on the same level as launch pads and gigafactories. From ground-based robots and self-driving cars to orbital data centers, every business line shares one underlying need — compute. Rather than waiting in line at external suppliers, why not own the entire supply chain?

Of course, $16.8 billion is only phase one. A 100-million-square-foot complex, accelerator-based lithography at scale, and the engineering challenge of the "world's largest building" all remain long roads. But regardless of whether Terafab hits its targets, it has already rewritten the definition of a chip factory — from a single plant into a fully vertical-integrated compute empire.

FAQ

Q1: When will Terafab start production, and how big is its capacity?

No official timeline has been announced. Phase one is a $16.8 billion investment, the complex exceeds 100 million square feet, and the long-term target is more than 1 terawatt of AI compute per year — roughly 50× current global AI chip output.

Q2: Will Terafab's in-house chips threaten NVIDIA?

Not in the short term — NVIDIA's software ecosystem (CUDA) and customer lock-in remain formidable moats. But if Terafab succeeds with accelerator-based lithography and supplies Tesla and SpaceX internally, it will gradually reduce their external procurement, potentially reshaping market supply and demand over time.

Q3: Why put design, fabrication, and packaging under one roof?

In the traditional model, design, foundry, and packaging are spread across companies and countries; each revision takes months. Co-locating everything lets engineers run a fast design→build→test→improve loop, shrinking iteration cycles from years to months or even weeks.

Q4: Will Terafab really use free electron laser (FEL) lithography?

Musk's "FEL FTW" reply to the ring-accelerator concept art shows the team is exploring accelerator-based lithography. But the technology is still early-stage; whether it can replace mature EUV remains to be proven, and it may be a long-term research track rather than the phase-one production line.

Q5: Is it reasonable to dedicate 75% of output to space data centers?

SpaceX's logic: orbital solar power isn't constrained by Earth's grid or weather, so satellite data centers could run 24/7. If Starlink-grade satellite networks solve the data-return problem, space compute is a genuine long-term answer to terrestrial power bottlenecks — though the engineering and cost challenges are extreme.