# Space-based AI data centers are racing to orbit and the economics are still deeply uncertain > Source: > Published: 2026-06-20 15:20:36+00:00 *A wave of well-funded startups is betting that solar-powered compute clusters in low Earth orbit will solve AI's energy crisis, but launch costs still run fifteen times higher than what makes the math work.* Starcloud became the fastest Y Combinator company in history to reach unicorn status, hitting a $1.1 billion valuation just 17 months after demo day on the back of a $170 million Series A. The Redmond, Washington startup launched its first satellite carrying an Nvidia H100 GPU in November 2025 and has already announced a partnership with Crusoe to deploy what would be the first public multi-tenant cloud in orbit, with limited GPU capacity available from early 2027. It has also received backing from Nvidia directly. For a company that existed in rough form less than two years ago, that is a remarkable trajectory. Then there is Baiju Bhatt, the Robinhood co-founder who quietly built a solar energy startup called Aetherflux before rebranding it Cowboy Space in May 2026 and raising $275 million at a $2 billion valuation. Bhatt's approach is more vertically integrated than anyone else in the field. After confronting the core bottleneck, the sheer lack of rockets capable of lifting the payloads he needed, he decided to build the rockets himself. Cowboy Space plans to integrate its orbital data centers directly into the second stage of its own launch vehicles: each upper stage becomes the satellite, weighing between 20,000 and 25,000 kilograms and delivering one megawatt of power to support nearly 800 GPUs. First launch is targeted before the end of 2028. Then there is Orbital, founded by Euwyn Poon, who sold e-scooter company Spin to Ford back in 2018 and has now turned to the stratosphere for his next act. Orbital raised $5 million in a pre-seed round from a16z's Speedrun accelerator in April 2026 and is focused specifically on AI inference rather than training, targeting a pathfinder demonstration mission in 2027 and a first full satellite in 2028. The Register noted that Orbital's own team has admitted the launch economics don't yet pencil out, which is worth filing away. The orbital compute thesis has a clean version of itself and a harder version. The clean version: no land acquisition, no power grid negotiations, no cooling infrastructure, and solar panels in low Earth orbit that can be up to eight times more productive than panels on the ground, generating power nearly continuously. Google's own Project Suncatcher, a research initiative announced in partnership with Planet Labs targeting two demonstration spacecraft around 2027, frames it the same way. Google is in talks with SpaceX to provide launch services for those satellites, which would eventually scale to 81 spacecraft spanning a one-kilometer orbital radius, each carrying Google's own TPU chips. The harder version comes from Wood Mackenzie, whose analysis found that orbital data centers currently cost roughly three to four times more than equivalent terrestrial infrastructure. One capital cost comparison for a 30.5-kilowatt deployment puts the space build at $3.1 million versus $382,000 on the ground. SemiAnalysis has mapped the trajectory: space compute reaches cost parity with terrestrial builds only around 2040, and only if launch costs fall to around $100 per kilogram, a fifteen-fold reduction from the current roughly $1,500 per kilogram. SpaceX's Falcon 9, which already represents one of the most dramatic launch cost reductions in history, still runs about $3,600 per kilogram to orbit. Starcloud's own projections are more optimistic: the company argues that operating a single 40-megawatt orbital cluster over ten years costs approximately $8.2 million, compared with $167 million for equivalent terrestrial infrastructure. That figure assumes the company can actually deploy and operate at scale, which no orbital data center company has yet done. Crusoe deploying its cloud platform on a Starcloud satellite scheduled for late 2026 will be the first real stress test of those numbers. ## The land-grab question The more interesting strategic question is not whether orbital compute becomes cheaper. It might, eventually. The real question is whether these startups are building defensible infrastructure businesses or whether they are demonstrating a concept that Google, Amazon, and Microsoft will absorb once the technical risk has been priced in by someone else's capital. Google's Project Suncatcher answers part of that question already. The hyperscalers are not waiting to see how Starcloud performs before deciding whether space-based compute is worth pursuing. They are building their own orbital roadmaps in parallel, and they have SpaceX relationships that no startup can easily replicate. Amazon's Project Kuiper is already a low Earth orbit constellation in its own right, and Microsoft has Azure Orbital for satellite connectivity. The infrastructure to extend both into compute in orbit is not a distant fantasy. Bhatt's decision to build his own rockets at Cowboy Space reflects a clear-eyed reading of this dynamic. If launch capacity is the constraint, and if SpaceX controls most of it while also servicing Google and potentially Amazon, then vertical integration is not an ideological preference but a survival strategy. The companies that depend entirely on third-party launch are exposed in ways that Cowboy Space, if it can actually deliver a rocket, would not be. What happens next probably depends less on who launches first and more on who lands the first major hyperscaler partnership or gets acquired by one. Starcloud's Nvidia backing gives it a relationship that matters enormously. Nvidia has every reason to want orbital compute to succeed, because orbital satellites running H100s and future Blackwell chips are just more GPU sales into a market that does not compete with its terrestrial customers. That alignment is more durable than venture capital enthusiasm. Whether Starcloud, Cowboy Space, and Orbital are building independent cloud businesses or very expensive acquisition targets is a question the next two years will answer. **Also read:** [Big Tech is borrowing like never before and the Fed just made that a lot more expensive](https://startupfortune.com/big-tech-is-borrowing-like-never-before-and-the-fed-just-made-that-a-lot-more-expensive/) • [You could be facing an AI chatbot at your next job interview](https://startupfortune.com/you-could-be-facing-an-ai-chatbot-at-your-next-job-interview/) • [Lloyds Banking Group bets £100 million on AI as it recruits the executive who rewired DBS](https://startupfortune.com/lloyds-banking-group-bets-100-million-on-ai-as-it-recruits-the-executive-who-rewired-dbs/)