Starcloud raises $170 million Collection Ato construct information facilities in house

Starcloud’s newest funding spherical values the house compute firm at $1.1 billion, making it one of many quickest startups to achieve unicorn standing after graduating from Y Combinator.

The corporate’s Collection A, which closed 17 months after its demo day presentation, was led by Benchmark and EQT Ventures. It’s one other signal of the curiosity in outsourcing information facilities to orbit as useful resource and political obstacles gradual their growth on Earth, however the enterprise mannequin relies upon on unproven know-how and important capital expenditure.

Starcloud has now raised a complete of $200 million, and launched its first satellite tv for pc with an Nvidia H100 GPU in November 2025. The corporate will launch a extra highly effective model, Starcloud 2, later this yr with a number of GPUs, together with an Nvidia Blackwell chip and an AWS server blade, in addition to a bitcoin mining laptop.

The corporate may also start growing an information middle spacecraft designed to launch from Starship, the reusable heavy carry rocket being constructed by Elon Musk’s SpaceX. Starcloud 3, because the spacecraft is called, can be a 200 kilowatts, three-ton spacecraft that matches the “pez dispenser” system SpaceX designed to deploy its Starlink satellites from Starship.

CEO and founder Philip Johnston mentioned he expects that would be the first orbital information middle that’s cost-competitive with terrestrial information facilities, with prices on the order of $.05 per kw/hour of energy — if business launch prices land round $500 per kilogram.

The problem is that Starship isn’t flying but; Johnston says he expects business entry to open up in 2028 and 2029. That’s the fact going through all the large house information middle tasks: highly effective house computer systems can be cost-prohibitive till a brand new era of rockets begins launching at a excessive operational cadence, one thing which may not occur till the 2030s.

“If it finally ends up being delayed, we’ll simply stick with it launching the smaller variations on Falcon 9,” Johnston mentioned. “We’re not going to be aggressive on vitality prices till Starship is flying incessantly.”

“There’s type of two enterprise fashions,” Johnston explains: One is promoting processing energy to different spacecraft on orbit; the corporate’s first satellite tv for pc, for instance, analyzes information collected by Capella Area’s radar spacecraft. Then, sooner or later when launch prices go down, extra highly effective distributed information facilities may doubtlessly pull work from their terrestrial counterparts.

That will get at how new this trade actually is. When Nvidia CEO Jensen Huang unveiled the corporate’s Vera Rubin Area-1 chip modules at his firm’s annual GPU Know-how Convention final week, he didn’t word that none had been produced or shared with the corporate’s growth companions. 

In actual fact, the variety of superior GPUs on orbit is numbered within the dozens, whereas Nvidia is estimated to have offered almost 4 million to terrestrial hyperscalers in 2025. 

Or think about that SpaceX’s Starlink communications community, the biggest satellite tv for pc community in orbit with 10,000 spacecraft, produces one thing round 200 megawatts of vitality, whereas information facilities with greater than 25 gigawatts of energy are at present beneath development within the U..S, in accordance with Cushman and Wakefield. 

Johnston argues that his firm is properly forward of the competitors, with the primary terrestrial GPU deployed in orbit. It was used to coach an AI mannequin in orbit, a primary, in accordance with Starcloud, and run a model of Gemini. Past the efficiency, Johnston says Starcloud now has priceless information about what it takes to run a strong chip in house.

“An H100 might be not the very best chip for house, to be sincere, however the purpose we did it’s we needed to show that we may run cutting-edge terrestrial chips in house,” he instructed TechCrunch. That tough-won information —one other GPU, an Nvidia A6000, failed throughout launch — will affect future designs.

There’s a laundry record of technical challenges to be solved, together with environment friendly energy era and cooling the hot-running chips. Starcloud-2 may have the biggest deployable radiator flown on a personal satellite tv for pc; he expects not less than two further variations of that spacecraft will head to orbit, Johnston mentioned.

Then there’s the problem of synchronization. The most important datacenter workloads, typically for coaching, require tons of or 1000’s of GPUs to work in tandem. Doing that in house will both require fantastically giant spacecraft, or highly effective and dependable laser hyperlinks between spacecraft flying in formation. Most firms engaged on this know-how count on these workloads to come back lengthy after less complicated inference duties happen on orbit.

Moreover Starcloud, Aetherflux, Google’s Venture Suncatcher, and Aethero — which launched Nvidia’s first space-based Jetson GPU in 2025 — are all growing house information middle companies. 

The elephant within the room is SpaceX itself, which has requested the U.S. authorities for permission to construct and function 1,000,000 satellites for distributed compute in house.

Going head-to-head with SpaceX is a frightening activity for any entrepreneur, however Johnston sees room for coexistence.

“They’re constructing for a barely completely different use case than us,” he instructed TechCrunch. “They’re primarily planning on serving Grok and Tesla workloads. It could be in some unspecified time in the future that they provide a 3rd social gathering cloud service, however what I believe they’re unlikely to do is what we’re doing [as] an vitality and infrastructure participant.”