A startup just raised $170 million to put data centers in orbit, and the math might actually work.

The Summary

  • Starcloud closed a $170 million Series A at $1.1 billion valuation to build solar-powered data centers in space, led by Benchmark and EQT Ventures
  • Space offers unlimited solar power and natural cooling, two things ground-based AI infrastructure is running out of
  • This isn't sci-fi anymore: launch costs have dropped 90% in a decade, making orbital compute economically viable

The Signal

The AI compute crunch just opened a new frontier. Starcloud's $170 million raise isn't about putting servers in space because it's cool. It's about solving two problems that are actively throttling AI development on the ground: power and heat.

Training frontier models already consumes gigawatts. Data centers are getting built faster than power grids can expand to feed them. In space, you get 24/7 solar with no atmospheric loss and radiative cooling that doesn't require water or HVAC. The physics work. The question was always cost.

That equation changed when SpaceX cut launch costs from $65,000 per kilogram to under $3,000. Suddenly, putting a rack of GPUs in orbit isn't insane. It's a capital allocation decision. You're trading launch cost for eliminating decades of utility bills and cooling infrastructure. For long-lived compute assets running continuous workloads, the amortization starts to pencil.

Benchmark leading this round matters. They don't chase science projects. They pattern-match to structural advantages. Space compute has three: infinite clean energy, passive cooling, and proximity to satellite networks that are becoming the backbone of global connectivity. That last part is key. If Starlink and its competitors are moving terabits per second overhead anyway, why pull that data back to Earth for processing?

The Implication

Watch who signs early contracts with Starcloud. The first customers won't be running consumer apps. They'll be training models, processing satellite imagery, or running inference for orbital applications. If this works, the agent economy gets a new tier: compute that never touches the ground. That changes cost structures, latency profiles, and who controls the infrastructure. The cloud was supposed to democratize compute. The orbital cloud might re-centralize it to whoever can afford the launch ticket.

Source: The Information