Starcloud is building a data center in orbit for artificial intelligence

Summary

NVIDIA-backed US startup Starcloud has raised $250 million at a $2.3 billion valuation to build GPU-equipped satellites for AI workloads in orbit. The company's ultimate goal is to provide 20 GW of computing capacity via 88,000 satellites.

Why it matters

AI's insatiable appetite for energy and land is pushing the sector to look beyond Earth for computing infrastructure. If Starcloud's orbital approach proves scalable, it could decouple AI growth from terrestrial power and cooling constraints. However, failure would serve as a high-profile reminder that space-based computing remains economically unviable.

Highlights

• Starcloud-1 successfully launched an NVIDIA H100 into orbit in November 2025 and ran Gemini inference in space.
• The ultimate goal: 88,000 satellites and 20 GW of capacity — an unimaginable scale-up from a single demo.
• Cooling suitable for space conditions, via radiators and radiation shields, remains an unsolved engineering challenge at the scale of satellite constellations.

Critical take

The leap from a single H100 in orbit to 88,000 satellites with 20 GW capacity is so extreme that the article's 'early stage' caveat feels like an understatement. Moreover, no comparative cost-per-FLOP or energy-per-token data is provided against terrestrial data centers, leaving the core economic thesis entirely unproven.