I just read about Terafab, and honestly, it’s one of the most ambitious projects I’ve seen in years. Elon Musk is presenting this as a hyper-scale chip factory that could completely change how we think about artificial intelligence computing.

What’s interesting is that Tesla, SpaceX, and xAI are working together on this. The goal is no small feat: to produce one terawatt of computing capacity annually, roughly 50 times more than the world currently produces in AI chips. To put it in perspective, the entire global AI capacity today is just 20 gigawatts per year.

Musk proposes integrating the entire cycle into a single facility. We’re talking about lithography masks, manufacturing, testing, and redesign all in the same location. That would dramatically accelerate improvement cycles compared to how the fragmented supply chain works now.

But here’s what surprises me: Musk argues that we can’t do this on Earth. He says that the total electricity generation in the U.S. is only about 0.5 terawatts, so physically, it’s impossible. Instead, he proposes AI satellites in orbit powered by solar energy. Prototype mini-satellites would need 100 kilowatts, but future versions could scale to megawatts.

To reach a full terawatt, they would need to launch about ten million tons of material into orbit annually. The current Starship V3 delivers 100 tons per launch, and V4 would double that to 200 tons. SpaceX has already completed over 500 successful booster landings.

What’s fascinating is the cost projection. Years ago, it was over $65,000 per kilogram. Now it’s between $1,000 and $2,000. If the optimized Starship achieves $100–200 per kilogram as Musk expects, space-based computing could be cheaper than terrestrial options within two or three years.

There are two types of chips in this initiative. One optimized for edge inference, designed for humanoid robots like Optimus and autonomous vehicles. Musk projects robot production could eventually reach one billion to ten billion units annually. The other type is designed to withstand space conditions, including high-energy particle bombardment.

The facility would start in Texas with state support. Musk acknowledges that existing suppliers remain critical but notes that their expansion rate is far below what Terafab would need. All the semiconductor plants on the planet combined account for just 2% of what they would need for their annual goal.

It’s ambitious, but if anyone can attempt something like this, it’s probably Elon Musk with the resources of these three companies. The combination of AI innovation, reusable rockets, and large-scale manufacturing is what makes this not just another futuristic idea.