Elon Musk announced TeraFab on Saturday in Austin: a joint venture between Tesla, SpaceX, and xAI to build a 100 million square foot chip factory targeting one terawatt of annual computing output at a cost of $25 billion. For scale, one terawatt exceeds total U.S. electricity generation. The facility has not been sited yet.
The architecture splits into two chips, one for Earth-based compute and one hardened for space deployment. Eighty percent of TeraFab output goes into AI satellites powered by solar, launched into orbit via SpaceX. The funding mechanism is a SpaceX IPO. The supply constraint driving all of this traces back to ASML, the single-source monopoly on extreme ultraviolet lithography machines.
The endgame described is a lunar electromagnetic mass driver that launches satellites into deep space without rockets. This episode is worth reading in full not for the conclusion but for the breakdown of how the two-chip architecture works, why the ASML bottleneck forced this vertical integration play, and whether the physics of a lunar mass driver actually holds up.
[WATCH ON YOUTUBE →]