Elon Musk, Jeff Bezos, and Google are pushing to put AI data centers in orbit, but physicists and investors like Sam Altman say the costs and engineering hurdles make little sense.
The pitch sounds almost too neat. Unlimited solar power, no real estate costs, no community opposition to massive power-hungry facilities. Just fire GPU-packed satellites into low Earth orbit and let the sun handle the rest. SpaceX CEO Elon Musk, Amazon founder Jeff Bezos, and Google CEO Sundar Pichai have all publicly backed the concept of orbital data centers as a way to feed AI’s voracious demand for compute without straining terrestrial grids. Musk put it bluntly on SpaceX’s website in February: terrestrial solutions alone cannot meet global electricity demand for AI without harming communities and the environment.
There is real money flowing behind this vision. Blue Origin recently filed with the US government to launch a constellation of 50,000 solar-powered data center satellites. Google’s Project Suncatcher is targeting a deployment in low Earth orbit by 2027. Nvidia has started shipping space-rated hardware, with CEO Jensen Huang telling reporters the economics will improve over time. Seattle-based startup Starcloud just hit a $1.1 billion valuation after a $170 million Series A, making it the fastest unicorn to emerge from Y Combinator. Robinhood cofounder Baiju Bhatt, whose space solar company Aetherflux is reportedly raising at a $2 billion valuation, has also placed bets in this category.
McKinsey and Company estimates the broader space economy could reach $1.8 trillion by 2035, a figure that venture capitalists are clearly taking seriously. The problem is that data centers specifically may be one of the least sensible things to put in orbit.
Matthew Buckley, a theoretical physicist at Rutgers University, frames the core issue with disarming simplicity. You physically could put a data center in space, he concedes. He just cannot see why anyone would. The scale of infrastructure required is staggering. Buckley estimates that a single orbital data center would need roughly 450 football fields worth of solar panels to generate enough power. That is not a satellite. That is a small country floating above the atmosphere.
Then there is the launch cost. Even as SpaceX drives down per-kilogram prices to orbit, putting enough hardware up to rival a terrestrial facility runs into tens of billions of dollars before you account for maintenance, cooling in a vacuum, radiation shielding for sensitive chips, and the latency of beaming data back to Earth. OpenAI CEO Sam Altman has called the entire concept ridiculous. Kathleen Curlee, a space economy research analyst at Georgetown’s Center for Security and Emerging Technology, describes it as a very wild idea, which in policy circles is not exactly a compliment.
Why the Idea Persists Anyway
For all its flaws, the space data center narrative solves a genuine problem that the industry has not yet cracked on the ground. AI training and inference demand enormous amounts of electricity, and utilities in the United States and Europe are already struggling to keep pace. Microsoft and Google have both seen their emissions rise as they scale data center capacity. Local communities from Virginia to Ireland are pushing back against new facilities that strain power grids and drain water supplies. The tension between AI ambition and physical infrastructure is real.
Orbital compute also carries strategic appeal. A data center in space operates beyond the jurisdiction of any single country, which could appeal to companies looking to navigate fragmented data sovereignty regulations. And for certain edge computing applications, having processing capability closer to satellite constellations could reduce the need to bounce data back to ground stations.
But those use cases are narrow, and the economics only work if launch costs continue to fall precipitously while solar panel efficiency rises dramatically at the same time. Neither trajectory is guaranteed on the timeline these companies are suggesting.
What makes this story worth watching is not whether orbital data centers become viable next year or even in five years. It is that some of the best-funded players in technology are treating them as inevitable. When Musk, Bezos, Pichai, and Huang all point in the same direction, capital follows whether the physics cooperates or not. The real question for startups and investors is simpler: are you backing the infrastructure that actually gets built, or the infrastructure that makes for the best keynote presentation?