Cowboy Space Corporation โ founded as Aetherflux in 2024 by Robinhood co-founder Baiju Bhatt โ closed a $275 million Series B this week at a $2 billion post-money valuation to put AI data centers in orbit. Index Ventures led the round, with Breakthrough Energy Ventures, Construct Capital, IVP, and SAIC participating; an earlier $80 million came from Index, BEV, Andreessen Horowitz, and NEA. The twist: Cowboy Space is also building its own rockets. Bhatt told TechCrunch he couldn't find enough commercial launch capacity from SpaceX or Blue Origin to scale a meaningful orbital data center business โ so he's building a launch vehicle in-house.
Each satellite is targeted at 20,000-25,000 kg, with 1 megawatt of solar power feeding roughly 800 onboard GPUs. The rocket would sit between Falcon 9 and Starship in lift class. Design choice: build the data center directly into the rocket's second stage rather than treat it as a separate payload โ a callback to Explorer 1, the first U.S. satellite, which was built as a rocket's final stage. Reusable booster planned eventually. Bhatt has hired Warren Lamont (former Blue Origin propulsion engineer) and Tyler Grinnell (former SpaceX launch director), and the company plans to build its own rocket engine โ the most complex and expensive part of any launch vehicle. First launch target: end of 2028. The thesis: solar irradiance in orbit is abundant and uninterrupted, radiative cooling sidesteps the water and grid-power constraints that are already throttling terrestrial data center expansion, and the bottleneck holding orbital back is launch cadence โ not satellites, not engineering, not demand.
Orbital AI data centers are now a category, not a one-off bet. Google's Suncatcher targets mid-2030s; Starcloud is starting on edge processing for space sensors; Anthropic just signed the Colossus 1 compute deal with SpaceX; and now Cowboy Space. The strategic bet across all of them: terrestrial data center power demand is hitting grid limits (40% of new data centers already in water-stressed areas per Xylem), and solar + radiative cooling in orbit removes both constraints. The pushback is real โ only SpaceX, Rocket Lab, and Arianespace are consistently launching commercial rockets; Blue Origin and ULA have been in development hell for years; Stoke Space, Firefly, and Relativity Space have been working a similar bring-up for years and still don't have operational systems. Bhatt himself admits the in-house-rocket move is "nuts," but the math is that satellites without launch capacity aren't a business โ and after talking to multiple providers he concluded "first-party rocket providers actually specialize into their own payloads," leaving outside customers underserved. The Aetherflux โ Cowboy Space pivot is also telling: the original plan (collect solar in orbit, beam it down) is harder than the new plan (collect solar, use it in orbit), and even the easier plan needs rockets they don't yet have.
First launch isn't until end of 2028 at earliest, which means the real technical credibility checkpoint is 2-3 years out. Intermediate milestones to watch: rocket engine bring-up, propulsion test stand, and whether Cowboy Space can absorb SpaceX and Blue Origin talent fast enough without inheriting the multi-year development-hell timelines of Stoke, Firefly, and Relativity. The framing for AI infra observers: orbital data centers aren't a hobbyist play, they're a deliberate bet that terrestrial scaling stops working in the next decade. If you believe AI compute demand keeps doubling โ and the QTS Georgia case and the Anthropic-SpaceX-Colossus deal both point that way โ the orbital math eventually wins. If you don't, $2B valuations for pre-revenue rocket companies look like froth. The personal capital signal matters: Robinhood made Bhatt rich, and his second act is "I'm going to build rockets and orbital GPUs." That's a macro bet against terrestrial AI's ability to grow into its power and water needs.