--- title: AI startup Orbital plans to launch its first data center into space in 2027 description: Los Angeles startup Orbital, backed by a16z Speedrun, plans a 2027 Falcon 9 launch of Orbital-1, a satellite housing NVIDIA Vera Rubin GPUs for AI inference in orbit. author: Darie Nani (Editor-in-Chief) date: 2026-04-15T11:15:47.397Z updated: 2026-04-15T11:15:47.408Z canonical: https://www.sovereignmagazine.com/article/orbital-ai-data-center-in-space-2027 image: https://cdn.nanimediahouse.com/orbital-satellite-1.webp.webp categories: Science & Tech content_type: Spotlight region: United States publication: Sovereign Magazine about: - type: Organization name: Orbital description: Orbital is a Los Angeles-based startup building and operating AI data centers in low Earth orbit. Each satellite houses a cluster of NVIDIA Space-1 Vera Rubin GPUs, powered by solar arrays and cooled by radiating heat directly into space. The first satellite, Orbital-1, is scheduled to launch on a SpaceX Falcon 9 in April 2027. Orbital was founded in 2026 by Euwyn Poon and is backed by a16z Speedrun. url: https://orbital.inc industry: Space Technology --- A new Los Angeles startup called Orbital is building AI data centers to run in low Earth orbit, and it now has a launch date. The company announced on Tuesday that it has raised funding from a16z Speedrun and will launch its first satellite, Orbital-1, on a [SpaceX Falcon 9](https://www.sovereignmagazine.com/article/spacex-acquires-xai-in-1-25-trillion-deal-ahead-of-record-ipo) in April 2027. Each satellite is designed to carry a cluster of NVIDIA Space-1 Vera Rubin GPUs, powered by solar arrays and cooled by radiating heat into space. Orbital is opening Factory-1, an R&D facility in Los Angeles, and is filing with the FCC for a full constellation. The first mission is framed as a validation flight: keep a server running in orbit, prove the radiation hardening holds, and then start selling AI inference as a commercial workload. The pitch is simple. Training a frontier AI model requires thousands of GPUs sitting next to each other, talking at near-zero latency, and that architecture cannot be smeared across satellites. Inference is different. Each query is independent and can be handled by any node with spare capacity. That is why Orbital is targeting inference rather than training, and why the founders think a satellite data center is a plausible product rather than a thought experiment. Behind the announcement is a thesis that has become hard to argue with. The bottleneck on AI is no longer chips. It is the [power required to run them](https://www.sovereignmagazine.com/article/ai-factories-are-the-new-data-centres) and the water and real estate required to cool them. Hyperscalers are signing nuclear power purchase agreements, restarting Three Mile Island, and queueing for grid interconnects that take five to seven years. The founders of Orbital have decided the cleanest way out of that queue is to leave it. ![Euwyn Poon, CEO and founder of Orbital](https://cdn.nanimediahouse.com/Euwyn.jpg) *Euwyn Poon, CEO and founder of Orbital* "AI progress is being constrained by the grid," said Euwyn Poon, Orbital's chief executive and founder. "Data center economics are dominated by electricity and cooling, and both are getting harder. In orbit, solar power is continuous and cooling is fundamentally different." In a sun-synchronous orbit a satellite sees the sun 24 hours a day with no weather and no night, and any waste heat can be dumped directly into the vacuum. Poon is not a typical space founder. He graduated from Cornell at 18, took a law degree, and spent two years as an M&A lawyer at Simpson Thacher before co-founding Spin, the bike and scooter sharing company that Ford bought in 2018. At Spin, he put hundreds of thousands of small electric vehicles on the streets of a hundred cities and scaled revenue past $100 million. After the exit he spent his time investing in AI infrastructure, and he says that is where he started to see the energy ceiling as a concrete constraint rather than a conference talking point. ## A new category is taking shape Orbital is entering a market that is beginning to take form. Starcloud, the Redmond startup formerly known as Lumen Orbit, launched a Starcloud-1 satellite in November 2025 carrying what it said was the first data-center-class NVIDIA H100 to run in orbit, and closed a $170 million Series A led by Benchmark and EQT Ventures in March. NVIDIA runs an orbital GPU program alongside Starcloud, and the first independent research groups are publishing on thermal and radiation design for on-orbit compute. The conversation has moved from whether this is possible to how fast it can scale. Orbital's bet is that inference demand will grow faster than any single operator can keep up with. The market is new enough to reward more than one serious builder, and Poon thinks a focused team with an operator founder and a dedicated R&D facility in Los Angeles can move quickly. A16z Speedrun writes early-stage checks designed to get companies to a first technical milestone, and Orbital-1 is that milestone. The chip is the piece that ties the category together. NVIDIA unveiled its Vera Rubin Space-1 module on March 17, a space-qualified variant of its Rubin platform which the company says can deliver up to 25 times the AI compute of an H100. That is the silicon both Orbital and Starcloud are now building around, and it is the most concrete sign that the major GPU supplier expects space compute to be a real market rather than a research curiosity. ## The engineering bet Orbital is taking on real engineering challenges, and the company knows it. Radiation is the most visible one. Consumer and data center GPUs that have flown to the International Space Station have survived but needed to be reset in flight, and a commercial service will need shielding, redundancy and recovery procedures that keep a cluster available around the clock. Orbital says the Orbital-1 mission is designed precisely to validate that hardware can run sustained workloads in orbit. Heat management is the other big one. A terrestrial data center [sheds heat into air and water](https://www.sovereignmagazine.com/article/the-swiss-startup-cooling-down-ai-s-21-billion-overheating-problem), and a satellite data center can only radiate it into space. That is part of why Orbital picked a sun-synchronous orbit and is designing the spacecraft around large radiator surfaces from the beginning rather than retrofitting a conventional server rack. It is a different starting point than a ground-based facility, and it is also the reason Poon's team sees an opening: the thermal design is not a workaround, it is the architecture. The company's focus on inference is a deliberate answer to the hardest constraint. Training needs thousands of GPUs coupled at near-zero latency, which does not map to satellites. Inference can be handled node by node and spread across a constellation, and the extra tens of milliseconds involved in reaching orbit and back sit inside the envelope that most modern AI products already operate in. ## What comes next The next milestone is Factory-1 coming online in Los Angeles, the FCC filing working its way through, and the Orbital-1 satellite getting built and integrated for a Falcon 9 flight in April 2027. If the test mission proves out the hardware, Orbital plans to start selling AI inference as a commercial workload and move toward a full constellation. The broader story is that the grid ceiling on AI has become the dominant constraint on scaling, and the industry is beginning to look for answers that do not depend on terrestrial power. Orbital is one of the first companies to put a launch date against that idea. The rest of the AI infrastructure industry will be watching, because if a satellite data center works at scale, the ceiling moves. **About Orbital** Orbital is a Los Angeles-based startup building and operating AI data centers in low Earth orbit. Each satellite houses a cluster of NVIDIA Space-1 Vera Rubin GPUs, powered by solar arrays and cooled by radiating heat directly into space. The first satellite, Orbital-1, is scheduled to launch on a SpaceX Falcon 9 in April 2027. Orbital was founded in 2026 by Euwyn Poon and is backed by a16z Speedrun. [Website](https://orbital.inc) ## FAQ **Q: Are there data centers in space?** Not yet at any commercial scale. Starcloud launched a satellite carrying a single NVIDIA H100 in November 2025, the first data-center-class GPU flown in orbit, but it is a test platform rather than a production service. Orbital plans to launch its first satellite in April 2027 with the same goal of validating GPU operations in orbit before selling compute commercially. **Q: How would a data center in space be powered?** By solar arrays. In a sun-synchronous low Earth orbit a satellite can see the sun around the clock with no weather and no night, so it can generate electricity continuously without relying on the power grid. Orbital's pitch leans on this plus radiative cooling, where waste heat is dumped directly into the vacuum instead of through air conditioning or water loops. **Q: What companies are building AI data centers in space?** Starcloud, based in Redmond, Washington, is the most advanced with an operational satellite and more than $200 million raised. Orbital, a Los Angeles startup founded by Spin veteran Euwyn Poon and backed by a16z Speedrun, announced its first test mission on Tuesday. Lonestar Data Holdings is also working on space-based data storage, but aimed at the Moon and focused on disaster recovery rather than AI workloads. **Q: Is NVIDIA building data centers in space?** NVIDIA is not operating its own space data centers, but in March 2026 it announced the Vera Rubin Space-1 module, a space-qualified version of its Rubin GPU platform. That is the chip Orbital and Starcloud are both designing their satellites around, which effectively makes NVIDIA the upstream supplier to the whole emerging category of the satellite data center.