Cowboy Space Raises $275M to Build AI Data Centers in Orbit

Baiju Bhatt's space company quietly rebranded this week from Aetherflux to Cowboy Space, raised $275 million in Series B at a $2 billion valuation, and announced an idea so audacious that it sounds either visionary or absurd: the only way to scale AI infrastructure past the wall of grid power and water is to put data centers in orbit. The company is not trying to compete with Equinix or AWS. It is trying to argue that those companies have a five-year ceiling.

What Actually Happened

Cowboy Space Corporation announced its $275 million Series B on May 11, 2026, led by Index Ventures with participation from new investors IVP, Blossom Capital, and SAIC, alongside existing backers Breakthrough Energy Ventures, Construct Capital, Andreessen Horowitz, NEA, and Interlagos. The round values the company at $2 billion. As part of the financing, the company unveiled the new Cowboy Space brand, having operated as Aetherflux since founding in 2024 with a narrower focus on space-to-Earth solar power beaming.

The new mission is broader and stranger. Cowboy is building a vertically integrated stack: a constellation of low Earth orbit satellites that harvest solar energy continuously above the atmosphere, a purpose-built launch vehicle to deploy them, and compute payloads designed for the unique radiation, thermal, and power conditions of space. The signature design choice is that the launch vehicle's upper stage stays in orbit and converts into a one-megawatt operational data center after delivering its payload. Each launch, in other words, builds the constellation while flying the mission.

The timeline is concrete. Cowboy plans to launch its first satellite later this year to demonstrate space-to-Earth power beaming, the original Aetherflux thesis. The first rocket launch is targeted before the end of 2028. The company is led by Baiju Bhatt, the billionaire co-founder of Robinhood, who pivoted personally into deep aerospace in 2024 after stepping back from the financial services world.

Why This Matters More Than People Think

The case for orbital compute is not a marketing story. It rests on three structural numbers the AI industry has spent the past year quietly trying to ignore. First, frontier model training capacity is now constrained by power, not by GPUs; Morgan Stanley estimates AI workloads will consume 1,100 terawatt-hours per year by 2028, a load the existing US grid cannot serve without major utility-scale buildouts. Second, water cooling is becoming a permitting nightmare in every major hyperscaler region. Third, the average permitting cycle for a one-gigawatt data center has stretched from roughly 24 months in 2022 to approximately 60 months today.

Orbit sidesteps all three. Solar irradiance in low Earth orbit is roughly twice ground-level intensity and runs around the clock. Vacuum cooling is effectively infinite if engineered correctly. There is no permitting authority. The catch is that getting compute to orbit, keeping it cold enough to actually compute rather than melt, and beaming results back through atmosphere is a stack of engineering problems that nobody has yet solved at scale. Cowboy's bet is that the economics get attractive somewhere between 2028 and 2032, and being the first integrated player at that moment is worth $2 billion of pre-revenue valuation today.

The Competitive Landscape

Cowboy is not alone in chasing this thesis. Lumen Orbit raised $11 million last year for orbital data centers. Starcloud has been demonstrating in-orbit GPU runs since late 2025. The European Space Agency has commissioned feasibility studies on lunar compute. China's Three-Body Computing Constellation launched its first batch of orbital compute satellites in 2025. The space-data-center category exists. What Cowboy claims is a structural advantage that other entrants lack: ownership of the launch vehicle.

That ownership matters more than it sounds. Every other orbital compute startup is dependent on SpaceX, Rocket Lab, or Blue Origin for launch capacity. Launch slots in 2027 and 2028 are already largely booked, especially for the Falcon 9 cadence that competitors need. By building its own rocket and architecting it so the upper stage becomes the data center, Cowboy reduces its launch costs to internal transfer pricing and turns the fundamental capacity bottleneck into an asset. If the integration works, the company captures most of the value chain from sunlight to served inference.

The competitive risk is also obvious. SpaceX, the most capable launch operator in the world, could announce its own orbital compute platform tomorrow and crush every dependent startup. xAI's hypothetical SpaceXAI merger discussion last quarter, with its rumored orbital data center roadmap, looks especially dangerous to Cowboy's positioning. The startup is racing to be vertically integrated before vertically integrated incumbents notice the category.

Hidden Insight: AI Just Became a Real Estate Problem

The non-obvious insight in the Cowboy story is what it tells you about how the AI industry is starting to think about itself. For the past three years, AI has been understood as a software and chip problem. Capex went into NVIDIA, training compute, and cloud. The most fashionable bottleneck conversation was about HBM memory, then about networking. In the past six months, the conversation has shifted unmistakably to power, then to land, then to permits. AI has become a real estate problem in disguise. Whoever controls the most physical sites with the most reliable power and cooling owns the choke point.

That shift is what a $2 billion valuation for a pre-revenue space startup is actually pricing. Investors are not betting that orbital data centers will displace Equinix in 2028. They are betting that the marginal terawatt of AI compute, the terawatt that nobody can permit on Earth, will be worth a premium that justifies the capital intensity of getting it there. If Morgan Stanley's 1,100 TWh forecast is even half right, the addressable opportunity for whoever cracks orbital compute is large enough to absorb every dollar Cowboy raises and then some.

The bear case, however, is not subtle. Putting computers in space sounds great until you remember that radiation hardens electronics into single-digit-percent yield problems, that thermal management with no convection is a graduate research challenge, that power beaming through Earth's atmosphere has efficiency losses that have killed every previous serious attempt, and that the unit economics of getting one server kilowatt to orbit and keeping it there for ten years remain unproven. Critics argue that Cowboy is selling a sci-fi narrative as if it were an engineering roadmap, and that the real winners in the AI infrastructure race will be terrestrial players who simply build faster, like Crusoe, Lambda, and the Saudi PIF-backed buildouts.

The risk is also that the timing will be wrong. If the AI capex cycle peaks in 2027 or 2028 and contracts thereafter, as some analysts now warn, the orbital compute thesis will arrive just as the appetite for it disappears. Building a constellation is not a pivot-friendly business model. The bet is binary, and the binary is not yet decided.

What to Watch Next

In the next 30 days, watch for the first Cowboy launch announcement. The company has indicated that the first satellite goes up later in 2026, demonstrating space-to-Earth power beaming. Whether that demonstration achieves better than 10 percent end-to-end efficiency is the question that determines whether the orbital thesis holds at all. Anything below 5 percent confirms the bear case; anything above 15 percent vindicates the bull case.

In 90 days, watch the customer pipeline. Cowboy is unlikely to land paid AI workload contracts before its 2028 launch, but customer-of-record letters of intent are the leading indicator. Anthropic, Microsoft, and the major hyperscalers have all publicly grappled with power constraints; an LOI from any of them would convert Cowboy from speculation to plausibility. Silence from hyperscalers will signal that the orbital thesis is still purely venture catnip.

In 180 days, the indicator to watch is SpaceX. If SpaceX announces an orbital compute initiative of its own by Q4 2026 or early 2027, Cowboy's window to establish itself before the dominant launcher enters the category will narrow sharply. If SpaceX continues to focus on Starlink and Mars, Cowboy retains its first-mover positioning and the $2 billion valuation looks defensible. The fate of this category is going to be decided as much by Elon Musk's calendar as by Cowboy's engineering.

AI's next bottleneck is not silicon or software; it is the power grid, and Cowboy Space is betting the grid quietly ends in low Earth orbit.

---