Starcloud secures $250M for orbital AI data centers

Starcloud has secured a $250 million extension to its $170 million Series A round from March, bringing its valuation to $2.3 billion. Led by Manhattan West Ventures, the round saw participation from Cisco, Benchmark, EQT, Soma, NFX, 776, Cedar Capital, Goanna Capital, Standard Capital, and Nvidia, which contributed $25 million. The 25-employee startup will use the capital to build a 100,000-square-foot manufacturing facility in Woodinville, Washington, and secure launch slots amid a tightening rocket market. With SpaceX planning to retire the Falcon 9 in 2028, Starcloud is racing to lock down transport for its massive planned constellation of up to 88,000 spacecraft.

The company plans to launch two of its next-generation 8 kW compute satellites, named Starcloud-2, on rideshare missions in 2027 to run orbital inference tasks for clients like U.S. government agencies. Ultimately, Starcloud is designing its largest spacecraft, Starcloud-3, to fly on SpaceX's massive Starship rocket. In the meantime, the startup is leveraging its position as the only known operator of a terrestrial Nvidia H100 GPU in orbit. Starcloud has already used this hardware to train an AI model in space, sharing the telemetry and thermal data with Nvidia.

For AI practitioners and hardware engineers, this partnership directly shapes the future of off-planet computing. Nvidia is using Starcloud's orbital data to develop its first dedicated space processor, the Vera Rubin Space-1 chip, which Starcloud hopes to launch by late 2028. Designing for space requires solving extreme engineering challenges, such as balancing chip running temperatures with radiator sizes, adding radiation shielding, and ruggedizing silicon to survive launch forces. If successful, Starcloud's orbital infrastructure could allow developers to run low-latency AI inference directly on satellite sensor data, bypassing the bandwidth bottlenecks of sending raw data back to Earth.