Elon Musk's Terafab: The $119 Billion AI Chip Factory

Kind: captions Language: en $119 billion for one chip factory. Could Elon Musk really build a chip factory so large that its planned manufacturing space reaches 100 million square ft, roughly 9.3 million m. The project starts with $16.8 billion, could eventually reach 119 billion, and is targeting enough computing capacity to reach 1 terowatt per year. Terrafab targets 1 terowatt of compute annually for Tesla, SpaceX, and XAI. It could reduce reliance on outside chip suppliers. The complex spans 100 plus million square ft, over three times Samsung's Pyongte campus. That makes Terraab more like a computing city than a normal chip factory. And there is an important detail here. 100 million square ft does not mean 100 million square ft of clean rooms. The number covers planned manufacturing space across the larger complex which is expected to include different parts of semiconductor production. SpaceX has described Terrafab as a vertically integrated facility covering manufacturing, packaging and testing of advanced logic and memory devices. The current plans show multiple enormous buildings rather than one simple rectangular factory. That makes sense because semiconductor manufacturing needs many different systems working together. Production areas, equipment rooms, packaging, testing, storage, and utilities all need their own carefully controlled spaces. At this scale, calling Terapab a factory almost feels like calling an airport a parking lot. Terrafab is not being designed to produce only one type of processor. The project is planned around advanced logic, memory, packaging, and testing, bringing several major stages together within the same industrial complex. That is important because modern AI hardware is not just about the processor itself. High performance systems also depend heavily on fast memory and advanced packaging. The first major production effort is expected to use Intel's 14A process technology which is still being developed. Musk has said the technology should be more mature by the time Terrafab begins production. Intel's involvement gives the project access to semiconductor manufacturing expertise that Tesla and SpaceX have not traditionally had. This is one of the most interesting parts of Terapab. Tesla already design some of its own chips, but designing a processor and manufacturing millions of advanced chips are two completely different jobs. A successful fab needs process engineers, equipment specialists, yield experts, and people who know how to move from laboratory results to high volume production. Earlier plans around Terraab described a target of scaling from roughly 100,000 wafer starts per month toward as many as 1 million wafer starts per month. That would mean multiplying monthly wafer capacity by 10 times. A wafer is a thin slice of semiconductor material on which many individual chips are manufactured. That does not mean one wafer equals one chip. A single wafer can contain hundreds or even thousands of individual dyes depending on the chip size. So moving toward 1 million wafers per month could represent an enormous number of individual processors. Although the final chip count would depend heavily on the size of each chip and the production process. This is also why Terrafab needs so much physical space. Semiconductor equipment is not packed together like shelves in a supermarket. Machines require carefully controlled environments, support systems, and room for material movement. The factory also needs chemical handling, water treatment, power infrastructure, and testing systems running alongside the production lines. The headline figure is $119 billion, but that needs some context. Tesla and SpaceX announced an initial investment of $16.8 billion for Terraab in August 2026. SpaceX's earlier filing described around $55 billion for an initial phase with total investment potentially reaching $119 billion if additional phases are completed. So $119 billion is not money being spent immediately. It represents the possible scale of the project if the full multi-phase plan goes ahead. Even the first $16.8 billion is enormous though. For a simple comparison, that is $16.8 $8 million for every $1 billion of the eventual headline figure. And the reason for the staged approach is obvious. Building a semiconductor factory this large all at once would be an enormous technical and financial risk. Starting with research and initial production allows the companies to test equipment, manufacturing processes, and supply chains before expanding further. Intel became a major part of the Terraab story in April 2026 when the company announced that it would work with Musk's Terraab project. Partnership is expected to support advanced processors for robotics and AI data centers while Intel's technology is also central to the planned manufacturing process. Musk has said Terrafab will use Intel's 14A process which makes the partnership much more important than simply supplying equipment. Intel has spent decades developing semiconductor manufacturing technology and that experience is difficult to recreate from zero. Building a fab is not like building another car factory where the basic manufacturing process can be copied and scaled. Tesla has also hired Gary Jen, a 17-year Intel veteran as a Terraab director. Jang's previous work included Intel's 18A manufacturing technology, factory construction, equipment installation, and the transition toward high volume manufacturing. That is exactly the type of experience a brand new semiconductor operation needs. One of the biggest challenges will be yield, which simply means the percentage of chips that actually work after manufacturing. If a factory produces 100 chips and only 70 pass testing, the yield is 70%. The other 30 chips consumed expensive materials, machine time, and energy without becoming usable products. At advanced process nodes, improving yield can take years. Tiny problems involving temperature, vibration, contamination, machine alignment, or chemical conditions can affect thousands of chips. And when the factory is producing millions of chips, even a small percentage of failures becomes a huge financial problem. This is why experience matters so much. Terrafab needs to solve two problems at once. Building enormous capacity and consistently producing working chips while its huge central circular structure has sparked speculation after Musk replied FFTW to a question about a possible particle accelerator. A free electron laser is real technology, but Terrafab has not officially confirmed that one will be installed there. The interest comes from its potential connection to EUV lithography, which uses 13.5 nanometer light today, although the exact purpose of the circular structure remains unconfirmed. Modern EUV lithography uses a laser to hit tiny molten tin droplets, creating plasma that produces 13.5 nanometer light for printing microscopic chip patterns. A free electron laser takes a different approach, accelerating electrons through magnets to produce highly controlled light. But FEL technology still needs to prove it can deliver the reliability, efficiency, and speed required for mass chip production. And Terapab faces another huge challenge. Keeping such a massive factory running with stable electricity 24/7. Because at this scale, even a short power problem could disrupt extremely expensive chipm equipment and production. And the timing is important. AI data centers are already pushing electricity demand higher across the United States. Terraab would be entering the same race for power except the electricity would be going into chip manufacturing rather than directly into AI servers. Water is another major requirement. Terapab's plans call for industrial water use from Gibbons Creek Reservoir rather than relying on local groundwater along with wastewater treatment, recycling, and conservation systems. So, the project is effectively building its own industrial ecosystem around the chip factory. The project is also expected to separate its manufacturing goals into two major areas. One is intended to produce computing hardware for Tesla vehicles and humanoid robots, while another is planned around high performance chips for space-based AI data centers. That division makes sense because the two workloads are very different. A vehicle processor needs to handle real-time driving tasks efficiently. While a data center processor may be designed around enormous AI workloads, one factory therefore has to support very different classes of computing hardware. Terrafab targets 1 terowatt of compute annually, connecting chips, robots, AI, and space systems across 100 million plus square feet and potentially 119 billion. The biggest challenge is producing advanced chips reliably at massive scale without creating new manufacturing bottlenecks. If you enjoyed the story, like, share, and subscribe for