Elon Musk wants to build a chip plant so vast it could churn out billions of AI processors every year, aiming for a computing capacity of one terawatt — fifty times more than the combined annual output of today’s biggest manufacturers.
The calm surface hides the scale — a facility aiming to outproduce TSMC and Samsung combined. Credit: New Atlas
The project is called Terafab, a joint venture between Tesla, SpaceX, and xAI. It begins in Austin, Texas, at Tesla’s Advanced Technology Fab, and Musk is pitching it as the largest end-to-end chip design and manufacturing facility ever attempted. Unlike existing fabs, which split design, lithography, fabrication, memory production, packaging, and testing across different sites and companies, Terafab would consolidate everything under one roof. That’s not just unusual — it’s unprecedented.
The scale Musk is talking about is staggering. Current industry leaders like TSMC and Samsung produce chips measured in gigawatts of compute power. Musk says Terafab will aim for one terawatt. To put that in plain terms: a gigawatt is roughly the output of a large nuclear power plant, while a terawatt is a thousand times that. He’s not talking about electricity generation, but about the raw computing capacity of chips designed for artificial intelligence. The facility is expected to cost upwards of $20 billion.
Two chip designs are planned. The first will power Tesla’s self-driving cars, robotaxis, and Optimus humanoid robots. Musk has repeatedly said he expects billions of humanoid robots to be needed in the near future — not just in homes but across industry. The second chip line is intended for space. SpaceX and xAI are working on orbital data centers, and Musk believes it will soon be cheaper to launch AI chips into orbit than to run them on Earth. Space-based compute would also allow direct use of solar energy, bypassing terrestrial infrastructure limits.
 |
| A quiet room today, but Musk claims it will soon house chips for orbital data centers. Credit: New Atlas |
Technically, Terafab is aiming for 2-nanometer process technology. A nanometer here refers to the size of features etched onto a chip; the smaller the number, the more transistors can fit, and the more efficient the chip becomes. Only a handful of companies in the world are even attempting 2-nm production, and none have yet achieved it at scale. Musk’s plan is to leap directly into that frontier.
The timeline is aggressive. Small batches are expected this year, with volume ramping up in 2027. Musk has floated figures of up to 10 billion robots annually, plus satellite data centers, all requiring chips. That means Terafab would need to produce several billion high-end AI chips every year. For comparison, the entire global semiconductor industry currently produces far fewer chips at this level of sophistication.
 |
| Austin’s skyline outside, but inside Musk is building toward orbit. Credit: New Atlas |
The obstacles are immense. Building a state-of-the-art fab requires cleanrooms, advanced air filtration, chemical handling systems, and years of process refinement. Even established players spend decades perfecting their methods. Musk is promising to do it faster, bigger, and more integrated than anyone before. As one industry analyst put it, “Constructing the facility alone takes years. Perfecting the architecture takes longer. Scaling it reliably is the hardest part.”
 |
| “One terawatt of compute power” — Musk’s words hang heavier than the machines behind him. Credit: New Atlas |
Skepticism is natural. Tesla’s Full Self-Driving technology has been “almost ready” for years, and the Hyperloop remains a concept. But Musk’s ventures have also delivered rockets that land themselves and cars that outsell incumbents. Terafab sits at that same intersection of audacity and uncertainty.
What’s clear is that Musk sees computing power as the bottleneck for everything from physics breakthroughs to interplanetary expansion. If Terafab succeeds, it won’t just be another chip plant. It will be the engine room for a future where AI runs cars, robots, and orbital data centers. If it fails, it will join the list of moonshots that never left the ground. Either way, the attempt itself shifts the conversation: the scale of ambition is now measured not in gigawatts, but in terawatts.
0Comments