The Redmond, Washington startup, which already has an Nvidia H100 GPU operating in orbit and has trained the first AI model in space, is now building a Starship-class spacecraft designed to be the first orbital data centre cost-competitive with terrestrial facilities.
Starcloud has raised $170 million in a Series A round led by Benchmark and EQT Ventures, valuing the Redmond, Washington, company at $1.1 billion. The round closed 17 months after Starcloud’s Y Combinator demo day presentation, making it one of the fastest startups in YC’s history to reach unicorn status.
Total funding is now $200 million. The company builds data centres in space, starting with GPU compute for other satellites and working toward a long-term vision of orbital infrastructure capable of handling the same workloads as terrestrial hyperscale facilities.
The core proposition is simple to describe and brutally difficult to execute. In orbit, solar power is essentially unlimited and free once a satellite is deployed. Cooling is passive: waste heat radiates into deep space, which runs at around -270°C, with no water required.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol' founder Boris, and some questionable AI art. It's free, every week, in your inbox. Sign up now!
There are no planning permissions, no grid connections, no land acquisition battles. Starcloud’s CEO and co-founder Philip Johnston argues that these structural advantages will make orbital data centres cost-competitive with terrestrial ones, once launch costs fall far enough. The problem, as Johnston openly acknowledges, is that the enabling technology is not yet operational.
Starcloud is nonetheless further along than any of its competitors. In November 2025, just 21 months after founding, it launched Starcloud-1: a 60kg satellite carrying an Nvidia H100 GPU, making it the most powerful GPU compute ever operated in space by approximately 100 times over.
The satellite subsequently became the first to train an AI model in orbit, specifically NanoGPT, trained on the complete works of Shakespeare, and the first to run a version of Gemini. The company is now processing data from Capella Space’s radar satellites in orbit, the first commercial use case. One Nvidia A6000 GPU failed during launch, a technical reality that will inform future hardware choices.
The Series A will fund three things. First, Starcloud-2, launching in October 2026: a more powerful satellite featuring multiple GPUs including an Nvidia Blackwell chip, an AWS server blade, and a bitcoin mining computer.
It will carry the largest deployable radiator ever flown on a private satellite. Second, the company will begin developing Starcloud-3: a 200-kilowatt, three-tonne spacecraft designed to fit the ‘pez dispenser’ deployment system SpaceX built for launching Starlink satellites from Starship.
Johnston says Starcloud-3 should be the first orbital data centre that can genuinely compete with terrestrial facilities on cost, with projected prices around $0.05 per kilowatt-hour, but only if commercial Starship launch costs reach approximately $500 per kilogram.
He expects commercial Starship access to open up in 2028 or 2029. If Starship is delayed, the company will continue launching smaller satellites on Falcon 9.
The strategic context is crowded and getting more so. SpaceX, which acquired Elon Musk’s AI company xAI in February 2026, has asked the US government for permission to build and operate a million-satellite distributed compute network in orbit. Blue Origin has expressed similar ambitions.
Google has Project Suncatcher; Aethero launched Nvidia’s first space-based Jetson GPU in 2025; Aetherflux recently reported raising at a $2 billion valuation.
The scale gap between what exists in orbit and what exists on Earth remains extraordinary: SpaceX’s Starlink network, with 10,000 satellites, generates around 200 megawatts of power, while data centres with over 25 gigawatts of combined capacity are currently under construction in the US alone.
The long-term plans, Starcloud’s includes an eventual 88,000-satellite constellation, assume a future that may not arrive on schedule.
Get the TNW newsletter
Get the most important tech news in your inbox each week.