Inside The Startup Launching AI Into Space

This feels like the birth of a new industry: data centers in space.

Earth data centers devour land, electricity, and fresh water, so Starcloud is starting small with a plan to scale into massive orbital compute that runs on sunlight and sheds heat into the cold of space.

They just launched a satellite carrying an Nvidia H100, the first data center‑class GPU to operate in orbit and roughly one hundred times the power of typical space computers.

We went to their Redmond HQ to hear how they moved from idea to on‑orbit demo in under two years.

What is Starcloud building?

We’re creating orbital data centers, first to sell GPU compute to satellites and then to compete on cost with the biggest facilities on Earth.

Paint the vision for running compute in orbit.

Anything that doesn’t demand ultra‑low latency can move to space, powered by continuous sunlight and cooled by radiating heat into deep space, which means zero fresh water use and less strain on the grid.

We’re aiming for forty‑megawatt modules of about a hundred tons, roughly what fits in a Starship payload bay.

How far along are you with the demo flight?

The H100 mission is to validate our thermal and radiation approach, then run real tests like high‑power inference, Gemini, fine‑tuning, and even training in space.

Why you, and why now?

Electricity demand is surging, especially from data centers, and our society is slow at building big infrastructure, so proving large, high‑compute satellites can help is an urgent bet.

We’ll scale by flying larger satellites that deliver meaningful compute for customers.

There’s been debate online; what’s the main critique you hear?

Can you respond to the claim that heat rejection makes this impractical?

The crux is radiator area, and that’s exactly our focus; my co‑founder with a PhD in engineering has spent a decade on deployable structures, and we’re building large, low‑mass radiators as core IP.

Does the skepticism motivate you?

Completely; there’s one hard problem—operate data centers in space cheaply—and if we crack it, everything else from hiring to fundraising gets easier.

You don’t come from traditional space hardware; what pushed you into this?

I started in software with a background in applied math and physics, always loved space, and the drop in launch costs made this feasible.

You first explored space‑based solar, right?

Beaming power down throws away about ninety‑five percent of it, which only works near fifty dollars per kilogram to orbit, so we pivoted to taking compute up, which pencils out around five hundred dollars per kilogram.

How did YC shape the plan?

We applied to serve satellites first and kept quiet about the bigger vision; YC pushed us to own it, even if saying most new data centers could be in space within a decade sounds bold.

Walk us through the build and team.

One co‑founder spent years on Microsoft data centers and SpaceX software and now leads our compute and radiation work; another is a veteran satellite designer with a PhD who handles structures and deployables.

We build the payload, power, and thermal hardware in‑house, run vibration and environmental tests, and partnered with Astro Digital for the bus; the first spacecraft is a sixty‑kilogram smallsat carrying the H100 and multiple antennas.

Launch day arrived.

SpaceX’s rideshare lifted off with Starcloud‑1 aboard to prove modern data center hardware can run in orbit.

The vehicle cleared the pad and the mission began nominally.

Over the next few months we’ll run a full test campaign, including the first on‑orbit model training and high‑demand inference.

What’s different about the second launch?

In October next year we’re flying a system that’s at least ten times more powerful with more GPUs on a next‑generation Nvidia architecture and always‑on optical links for high bandwidth and low latency.

The five‑gigawatt dream is likely a decade away, but this first step is real, and others are now looking up.

Google, SpaceX, and Amazon are exploring orbital data centers that run on the sun and vent heat into space.

Anything worth doing is hard, so we chose the most ambitious path we could: build the future where most new data centers are in orbit.