Orbital

AI compute workloads are expanding faster than Earth's power infrastructure can support, with data centers projected to need double today's power by 2030. The text states that every gigawatt of new data center capacity today is a fight for power, land, and permits, and that there is no easy path forward on Earth using existing grid infrastructure, which the company describes as century old.

Orbital is building a constellation of satellites in Low Earth Orbit that capture solar energy to power AI compute hardware directly in space. In LEO, the sun delivers 1,361 W/m\u00b2 continuously with no atmosphere, weather, or night, which the company says is more than 5x the energy density of ground-based solar. Space also serves as a passive heat sink near absolute zero, allowing radiative cooling of servers at zero operating cost. The roadmap starts with a 2027 pathfinder mission flying a GPU module as a hosted payload on a SpaceX Falcon 9 to validate radiative shielding and thermal architecture, followed by Orbital-1 in 2028, a purpose-built satellite with multiple GPU nodes, high-bandwidth ground links, and commercial inference availability. The company has also filed plans for a 100,000-satellite constellation.

The page argues data centers will need double today's power by 2030 and that grid power, land, and permitting on Earth are already major bottlenecks, framing space as a way to remove that constraint.