Starcloud

Starcloud puts data-centre GPUs on satellites. Starcloud-1, launched in November 2025, carries an Nvidia H100, the chip terrestrial AI clusters run on, and the company says it is the only one operating such a GPU in orbit and the first to train a model there. The argument is that orbit offers continuous sunlight for power, radiative cooling for heat and no permits to fight, so compute can scale toward gigawatts without the constraints a data centre faces on the ground.

Philip Johnston, Ezra Feilden and Adi Oltean founded the company in 2024 and went through Y Combinator that summer. Feilden designed satellites at Airbus and Oxford Space Systems, and Oltean built Starlink beam tracking at SpaceX. The money has come fast. A $170 million Series A closed in March 2026 and a $250 million extension in August, led by Manhattan West with $25 million from Nvidia, at a $2.3 billion valuation, and the company has asked the FCC for permission to operate 88,000 spacecraft.

One satellite with one GPU is a demonstration rather than a data centre, and the plan runs through two more generations, Starcloud-2 compute satellites of about 8 kilowatts on rideshares in 2027 and Starcloud-3, a far larger spacecraft sized for Starship. The constraint the company itself names is launch, which it calls constrained now and which gets no easier if Falcon 9 stops flying in 2028 as SpaceX has said, so the larger spacecraft wait on Starship.