SpaceX plans more than $74 billion for two AI supercompute campuses in Texas and Arizona

SpaceX wants to own its training ground, not rent it. On 12 November the company said it plans to spend more than $74 billion on two AI supercompute campuses, and that the compute will train its own super intelligence models.

In separate news, NVIDIA said in a filing on 12 November that Jensen Huang resigned as its President, Chief Executive Officer and a member of its board, and that the board accepted his resignation on 11 November.

Think of a team announcing two new stadiums at one press conference. One is already going up: SpaceX plans more than $24 billion at Starbase, Texas, says the project has already generated more than 3,000 construction jobs, and says the site will make its own electricity instead of drawing from the public grid. The other is only proposed so far: a roughly $50 billion campus in Cochise County, Arizona, with nine compute buildings planned for up to six gigawatts of power, which SpaceX is still seeking approval to build.

What happened is a company statement of planned spending. What could happen is a multi-gigawatt training base on the ground. What is still a guess is timing, because SpaceX gives no start date for either campus and names no chip supplier.

SpaceX is in our AI Economy Index, in the Major division. We file this as a capacity plan, with no new link on the map and no colour change.

What would change this read: a local approval or rejection in Arizona, a named chip or power partner, or SpaceX reporting installed capacity at either site.

This is general information, not investment advice.