The economics of AI in orbit: why space data centers cost three times as much as ones on Earth

# The economics of AI in orbit: why space data centers cost three times as much as ones on Earth

Moving AI compute beyond Earth's atmosphere sounds appealing: unlimited solar power, ideal conditions for cooling hardware, no fights over land. But once engineers start running the numbers, the romantic idea collides with the cold math of economics. Building a 1-gigawatt orbital data center would cost roughly $42.4 billion — almost three times as much as its terrestrial counterpart. Those figures call into question whether space-based AI infrastructure makes sense at all, and they explain why leading tech companies keep pouring trillions into capacity on the ground.

Space logistics are unforgiving. Getting hardware into orbit is a problem of an entirely different order than building a data center in the Nevada desert. Every kilogram of equipment needs a rocket, fuel, and launch hours. Current commercial rockets such as Falcon Heavy cost about $62 million per launch, and the payload tops out at 60 tons. Even assuming launch costs fall in the future, logistics alone will eat a substantial share of the budget. Add frequent repair missions, replacement of failed hardware, and a constant supply of spare components, and it becomes clear why the estimate grows exponentially.

In space, every engineering problem demands not just an expensive solution but an extremely expensive one. Vacuum and extreme radiation degrade electronics much faster than on Earth. Servers need extra shielding and protection, which adds mass and cost. The cooling system doesn't need water the way terrestrial data centers do, but it does require costly radiators and thermal control to dissipate heat into the vacuum. Every screw, wire, and chip has to be tested for resilience to space conditions, which multiplies component costs several times over. Redundancy and safety systems also become critical when you can't simply send a technician out for a quick fix.

Even orbit's supposed advantages look less attractive under closer scrutiny. Yes, solar panels in orbit receive an almost constant stream of energy, but solar arrays in space degrade from radiation. Cooling really is simpler thanks to radiating heat into the vacuum, but supercomputer-scale hardware produces so much heat that the advantage disappears. Meanwhile, data centers on Earth keep getting more efficient. Companies have already reached power consumption of 1.2 kilowatts per processor, and innovations in liquid cooling and materials keep improving the numbers.

Ground-based alternatives are advancing faster and more cheaply. Building a terrestrial data center of the same capacity would cost about $14 billion, and that figure could fall thanks to scale and competition. Companies can place servers in countries with cheap electricity, cheap labor, and climates that favor cooling. Iceland, Norway, Canada, and Middle Eastern countries with surplus electricity — that's where the real investment is going.

Orbital AI infrastructure will remain an attractive idea for textbooks and research papers, but for the foreseeable future the economics simply won't let companies turn it into reality. Earth is still far cheaper than space.