TLDR:
- Terafab will produce two chip types; one for Tesla and Optimus, and a space-hardened D3 variant for orbit.
- Solar panels in space run five times more efficiently, making orbital AI cheaper to operate than ground-based systems.
- A lunar electromagnetic mass driver could slash payload launch costs from $1,200 per pound to just dollars in electricity.
- One entity now controls the rockets, chips, robots, and satellites needed to build an off-planet AI supply chain.
Terafab, a semiconductor facility developed by Tesla, SpaceX, and xAI, has officially broken ground. Elon Musk unveiled the project Saturday night at a decommissioned power plant in Austin, Texas.
The facility targets one terawatt of AI compute annually, roughly double the total electricity capacity of the United States.
Around 80% of its chip output is set for space deployment. Musk framed the effort as the start of what he called a galactic civilization.
Terafab’s Chip Strategy and Space-Bound AI Infrastructure
Terafab will produce two distinct types of chips. One type supports Optimus robots and Tesla vehicles. The other, designated D3, is hardened specifically for space.
Most of the facility’s output, roughly 80%, is directed toward orbital deployment. The remainder supports ground-based AI applications and consumer devices.
Musk expects Optimus robot production to reach 10 to 100 times the volume of car manufacturing. That points to billions of chips being produced annually.
The scale makes Terafab central to both commercial and space operations. No existing facility currently targets this combined level of output.
Musk told the Austin audience that solar panels in space operate five times more efficiently than on Earth. Milk Road AI reported this as a central part of its cost argument for orbital AI.
Space also provides uninterrupted sunlight, unlike ground-based installations. Over time, this positions orbital AI as cheaper to run than terrestrial alternatives.
Near-term chip output from Terafab is directed toward a data center under construction in Virginia. That facility serves as the initial hub before full orbital deployment begins.
It connects ground-level production to the broader space strategy. From there, the roadmap extends outward toward the moon.
Lunar Mass Driver and the Road to a Petawatt
Beyond the terawatt lies a petawatt target, one thousand times more powerful. Musk argued that reaching it requires moving manufacturing off-planet.
The moon, with its low gravity and no atmosphere, becomes the logical production site. A lunar base forms the next stage of the infrastructure plan.
Rather than rockets, the plan calls for an electromagnetic mass driver on the lunar surface. This magnetic cannon would launch AI satellites directly into deep space.
A Falcon rocket currently costs around $1,200 per pound of payload. A lunar mass driver could reduce that figure to just dollars per pound in electricity.
Milk Road AI described this as potentially the single biggest reduction in the cost of intelligence in human history, with the caveat that it must first work.
That qualifier is worth noting. No mass driver of this scale has been built or tested. The engineering challenges ahead remain unresolved.
Musk stated his goal to complete the lunar infrastructure within his own lifetime. Terafab has already broken ground, and the D3 chips are currently in design.
The race to place AI infrastructure in space has formally started. One entity now controls the rockets, the robots, the chips, and the satellites required to pursue it.
