Google is in talks with SpaceX to launch computing facilities into orbit. This is not science fiction. Google has already been working on the idea since late 2025 under a project called Suncatcher, has partnered with a satellite builder called Planet Labs to design the hardware, and plans to launch two test satellites by early 2027. SpaceX wants to be the rocket company that puts them up there.
But Google is also talking to other launch companies at the same time. That detail matters. It tells you that for Google, this is a supplier negotiation, not a partnership. SpaceX is just one bidder.
The real story here is not about space. It is about a power crisis that is already slowing down AI on Earth. Right now, nearly half of all planned AI data center projects globally are delayed, almost entirely because there is not enough electricity to run them. In parts of the US, the wait just for the electrical transformer equipment needed to connect a new data center to the grid now stretches beyond three years. Building a new power transmission line from scratch takes seven to ten years. Northern Virginia, which hosts the world's largest concentration of data centers, has effectively stopped approving new ones in several counties until the power infrastructure catches up.
A solar-powered satellite in space sidesteps every one of those problems. Satellites in certain orbits receive sunlight almost continuously, generating power at roughly five times the efficiency of the same solar panels on Earth. There are no permits, no angry local residents, no grid connections to wait for, and no power bills. The appeal is genuine.
The economics, however, are still deeply unfavorable today. Independent engineering analysis puts the total cost of a space-based computing network at around $51 billion, compared to roughly $16 billion for an equivalent ground-based setup over the same five-year period. That is about three times the cost. Google's own research team acknowledges this honestly, noting that launch costs would need to fall to under $200 per kilogram before orbital data centers become competitive, and projecting that this may not happen until the mid-2030s. Right now, launch costs are still multiples above that threshold.
Then there are the technical problems that rarely make the headlines. Computer chips in space are bombarded by radiation constantly, which degrades their reliability. Cooling, which is easy on Earth using water and air, has to be done in a vacuum using large physical radiators. Those radiators add weight, which adds launch cost. And if hardware breaks down, there is nobody up there to fix it.
So why is this happening now? The SpaceX IPO. SpaceX is heading toward what would likely be the largest public stock offering in history, targeting a valuation of around $1.75 trillion this summer. Orbital data centers have become a central part of that pitch to investors, with the promise that space-based computing will be cheaper than ground-based within just a few years. Signing up Google as a customer before the roadshow is an extremely valuable thing for SpaceX to be able to announce. The same logic applied last week when Anthropic agreed to use SpaceX's Memphis data center, with a stated interest in space-based facilities in the future.
Elon Musk also now controls both the launch company and an AI company, after merging SpaceX with xAI in February. That makes orbital data centers a closed loop where SpaceX launches, SpaceX's AI products consume the compute, and SpaceX collects revenue on both ends. The incentive to make this story credible before the IPO is enormous.
For the rest of the world, the more grounded implication is this: the scramble for power is reshaping where and how AI computing gets built. Companies are moving to regions with available electricity, signing long-term nuclear power deals, building their own gas plants, and now seriously exploring space. Any business that depends on AI-powered services should expect the cost and availability of those services to reflect these infrastructure constraints over the next several years. The power problem is real. The space solution is a decade away from being affordable.