<cite index="81-1">During SpaceX's first earnings call as a publicly traded company, Musk announced the company will build exclusively on Nvidia's next-generation Vera Rubin architecture, calling it the best AI computer</cite>. <cite index="89-2">SpaceX announced an exclusive commitment to build its future compute capacity on Nvidia's Vera Rubin architecture at unprecedented scale: 2 gigawatts (GW) of installed compute by the end of 2026, scaling to approximately 10 GW by the end of 2027, with a tentative 20 GW power and cooling infrastructure target</cite>.
<cite index="83-4">SpaceX booked $18.37 billion of capital expenditure for Q2 2026, with $15.83 billion of it going to AI infrastructure, against a consensus estimate closer to $13.22 billion—a rocket company spent about twice its quarterly revenue on computing hardware in three months</cite>. <cite index="86-2">SpaceX ended the quarter with 1.4 gigawatts of nameplate compute capacity and expects to exceed 2 GW by year-end 2026, with 2027 ambitions ranging from roughly 10 GW to as much as 20 GW of power-and-cooling capacity</cite>.
<cite index="86-2,86-3">SpaceX and Nvidia announced a joint program to design the Starmind AI1 satellite compute payload, with each Starmind satellite housing Rubin GPUs and Vera CPUs, and plans to deploy Nvidia's NVL72 rack-scale systems on the ground and in orbit</cite>. <cite index="82-3">SpaceX expects to begin launching the first Starmind AI satellite next year, an optimized version of the Vera Rubin NVL72 computing system, with the company planning to use the same AI architecture both on Earth and in orbit</cite>.
<cite index="86-3">SpaceX's contracts tied to Anthropic's Colossus 1 involve roughly 300 megawatts of capacity, about 220,000 Nvidia GPUs, and payments near $1.25 billion per month through May 2029</cite>. <cite index="89-3">According to Melius Research analyst Ben Reitzes, even the initial 2 GW deployment could add roughly $100 billion in incremental revenue for Nvidia</cite>.
<cite index="87-3">SpaceX views the NVL72 design as a radical simplification of the standard rack, deploying it on the ground as well as in orbit—the same chips, same racks, same software stack, different address—which is how infrastructure transitions actually happen through standardization that can flow toward wherever power and cooling are cheapest</cite>. <cite index="86-3">SpaceX has filed with the U.S. Federal Communications Commission for authority to deploy up to one million orbital data-center satellites between 500 and 2,000 kilometers altitude, describing the planned constellation as a 'petabit laser mesh'</cite>.