Starcloud builds orbital computing satellites that process data in space and aim to grow into full-blown off-world cloud infrastructure. Its latest Series A values the company at $1.1 billion, a striking number for a startup pitching space data centers just 17 months after Y Combinator demo day. The pitch is simple: Earth is running into power, land, and political constraints as AI data centers scale. Philip Johnston, Ezra Feilden and Adi Oltean founded Starcloud in 2024 in Redmond, Washington.
What is Starcloud and how do its space data centers work?
Starcloud’s product is basically orbital compute infrastructure. A customer with a satellite or space station can send raw data to Starcloud’s spacecraft and run GPU-heavy processing in orbit. It can store results there, then send down smaller, more useful outputs instead of dumping huge raw files back to Earth. Starcloud-2 delivers this through a smallsat with a GPU cluster, persistent storage, 24/7 access, and custom thermal and power systems. The company plans to make it fully operational in sun-synchronous orbit by 2027.
Starcloud already has its first proof point in orbit. Starcloud launched Starcloud-1 in November 2025 with the first Nvidia H100 GPU in orbit. In December 2025, that satellite ran a version of Gemini in space and became the first spacecraft to train an LLM in orbit using nanoGPT. That matters less as a stunt than as a hardware test: if a terrestrial AI chip can survive launch and operate in space, the roadmap stops sounding totally ridiculous.
For spacecraft operators, the before-and-after is pretty clear. Before Starcloud, an Earth observation company often has to downlink massive data sets to ground stations, wait for bandwidth, then process the payload on Earth. With Starcloud, the bet is that a lot of that work gets done in orbit first. That cuts latency and avoids wasting bandwidth on raw data. That’s why Starcloud’s first satellite is already analyzing data from Capella Space’s radar spacecraft, and why the company talks about satellite data processing as its first real business rather than chasing giant AI training jobs on day 1.
There’s a second customer type built into the plan. Starcloud-2 isn’t only pitched to in-space users. It’s also framed as a secure, sovereign cloud node for terrestrial users who want storage and compute that sit outside any single country’s physical infrastructure. That’s niche today. Still, it shows where Starcloud wants to go: not just edge compute for satellites, but orbital data centers that can eventually pull workloads away from Earth.
Who founded Starcloud and why build space data centers?
The founding story
Johnston’s pitch has always been blunt: if AI keeps scaling, Earth-bound data centers will slam into energy and permitting limits. Starcloud was built around that thesis in 2024, first with smaller orbital compute missions and then with a longer-term plan to launch much larger platforms once heavy-lift economics improve. Y Combinator’s profile shows the company was already designing a “micro data center” for a 2026 launch and a larger “Hypercluster” for the Starship era.
Why this team has unusual founder-market fit
Johnston isn’t a random AI founder trying on a space startup costume. He’s a second-time founder who previously worked at McKinsey on satellite projects for national space agencies, and he has degrees from Harvard, Wharton, and Columbia, plus a CFA charter. It’s an odd résumé for a rocket-adjacent company. But that mix of aerospace, policy, and capital markets fits a business that lives or dies on both engineering and launch economics.
Feilden brings the spacecraft side. He spent about a decade in satellite design, with work at Airbus Defence & Space and Oxford Space Systems, including missions tied to NASA’s Lunar Pathfinder. His background in deployable solar arrays and large structures is especially relevant because Starcloud’s long-term design problem isn’t just compute. It’s also power generation and thermal management.
Oltean rounds out the compute piece. At SpaceX, he worked on Starlink networking for in-motion use cases, including Starship-related work. Before that, he worked at Microsoft on large GPU production clusters and early LLM infrastructure, and Y Combinator says he holds more than 25 patents. That’s the kind of operator background investors like because Starcloud is trying to merge spacecraft engineering with data center engineering. Not replace one with the other.
Early execution, fundraising, and the real competition
Starcloud has moved fast enough to make investors tolerate a very capital-intensive story. The company has now raised $200 million in total. Benchmark and EQT Ventures led this Series A, and it closed just 17 months after demo day. Starcloud had already launched Starcloud-1, booked follow-on missions, and secured LOIs for H100 compute time in space before landing this round. It also set up payload manufacturing in Redmond. This isn’t a slide-deck company anymore.
The competition is getting real, though it’s still messy. Aetherflux, founded by Robinhood co-founder Baiju Bhatt, has shifted from laser power transmission toward powering space data centers and expects its first data center satellite in 2027. Aethero is coming at the category from a different angle. It’s building radiation-hardened edge computers for satellites; its Jetson-based NxN module was built to hit 100 TOPS in a CubeSat-compatible form factor.
Starcloud’s edge is simple: it has already put a terrestrial H100 in orbit, and it’s talking about commercial workloads now rather than only future architectures. The legacy alternative is still Earth. And that’s a massive incumbent. Investors aren’t backing Starcloud because space is easier. They’re backing it because if launch costs fall enough, energy economics could flip.
Why does Starcloud’s Series A matter?
This round matters because Starcloud isn’t raising money to polish software. It’s raising to build hardware that only starts to make sense at scale.
Later in 2026, the company plans to launch Starcloud-2 with multiple GPUs, including an Nvidia Blackwell chip and an AWS server blade. It’ll also carry a bitcoin mining computer. That sounds a little chaotic. It’s also smart. Starcloud needs data on power, cooling, fault tolerance, and mixed workloads in orbit, not just one clean demo.
Then comes Starcloud-3. The company is designing that spacecraft to be a 200-kilowatt, 3-ton orbital data center designed for SpaceX’s Starship deployment architecture — the “PEZ dispenser” system built for Starlink satellites. Johnston thinks that vehicle could be the first one to compete with terrestrial energy costs, with power on the order of $0.05 per kilowatt-hour if launch prices land around $500 per kilogram.
That “if” is doing a lot of work.
Starship still isn’t in commercial service, and Johnston has said he expects access to open in 2028 and 2029. He’s also been candid that Starcloud won’t be competitive on energy cost until Starship is flying frequently. If that slips, the fallback is to keep launching smaller versions on Falcon 9. So this Series A is really a bet on two roadmaps at once: Starcloud’s spacecraft roadmap and SpaceX’s launch roadmap.
Johnston also broke the business model into 2 parts. Near term, Starcloud sells processing power to other spacecraft. Longer term, if launch gets cheap enough, it wants distributed orbital clusters to pull work from Earth-based data centers. Investors didn’t just fund a satellite company here. They funded a staged transition plan.
What’s driving demand for space data centers now?
The terrestrial data center market is a huge part of the story. In the Americas alone, operational data center capacity has reached 43.4 GW, with another 25.3 GW under construction, and nearly 89% of that pipeline was already pre-committed in early 2026. Vacancy was just 4.2%. In plain English: the market is building like crazy and still looks tight.
It’s not just demand. It’s where and how you’re allowed to build. Cushman & Wakefield points to power availability, grid access, permitting friction, land use, and local regulation as the bottlenecks shaping the next wave of projects. That lines up almost perfectly with Starcloud’s pitch that resource and political obstacles on Earth create room for in-space computing.
There’s still a giant reality check, though. Only dozens of advanced GPUs currently operate in orbit, while Nvidia sold nearly 4 million chips to terrestrial hyperscalers in 2025. SpaceX’s Starlink network may be the biggest satellite constellation ever built, but even that produces only around 200 megawatts of power. Meanwhile, terrestrial facilities with more than 25 gigawatts of capacity are under construction in the U.S. alone. The scale gap is absurd right now.
That’s why the interesting question isn’t whether orbital data centers replace terrestrial ones soon. They won’t. The question is whether some high-value workloads — especially Earth observation, defense, sovereign compute, and latency-sensitive edge inference — migrate first. If they do, Starcloud doesn’t need to win the whole market to build a very real business.
Should you take Starcloud’s orbital data center plan seriously?
Honestly? Yes — but only if you treat it as a long-duration infrastructure bet, not a normal startup growth story.
Starcloud has already done the part that many deep-tech companies never do: it put hardware in orbit and ran modern AI silicon there. It turned a weird thesis into something concrete. That’s a big deal. So is the fact that Benchmark and EQT were willing to fund it at a $1.1 billion valuation.
But the company still depends on breakthroughs that aren’t fully under its control. Launch cadence. Launch cost. Space-rated cooling. Multi-satellite synchronization. All of that has to work before space data centers look anything like a mainstream cloud category.
So the next thing to watch isn’t the valuation. It’s whether Starcloud-2 launches in 2026 and whether Starcloud can turn orbital compute from a technical flex into repeatable revenue.
Read how Qodo raises $70M for AI code review verification as it builds a system to check and secure AI-written code before deployment
FAQ
What is Starcloud’s latest funding round?
Starcloud’s latest round is a Series A that values the company at $1.1 billion. Benchmark and EQT Ventures led the financing, and the company has now raised $200 million in total after reaching that milestone only 17 months after Y Combinator demo day.
How do Starcloud’s space data centers actually work?
They work by putting GPU-equipped satellites in orbit so data can be processed before it ever comes back to Earth. Starcloud-2 is designed with a GPU cluster, persistent storage, always-on access, and custom thermal and power systems, while Starcloud-1 already proved an H100 could run AI workloads in space.
Who founded Starcloud?
Starcloud was founded in 2024 by Philip Johnston, Ezra Feilden, and Adi Oltean in Redmond, Washington. Johnston came from McKinsey satellite work, Feilden from Airbus and Oxford Space Systems, and Oltean from SpaceX and Microsoft’s large-scale compute infrastructure world.
Is Starcloud in the cloud market or the space market?
It’s really both. Starcloud sits in the emerging orbital data center category — part satellite infrastructure, part cloud computing, part edge AI — and its near-term focus is selling in-space compute to spacecraft operators before chasing broader terrestrial workloads.
