Elon Musk says AI computing could become cheaper in space within the next few years. The idea is to launch solar power satellites together with AI chips and run the calculations right where the sunlight falls. Several companies and researchers, including Google's Project Suncatcher, are also exploring orbital computing.
Musk's timetable runs a few stops ahead of reality. Read his words as prophecy, and they are often wrong. Read them as a purchase order for industry, and they become useful. “Computing in space will be the cheapest in a few years” is not a report on finished technology. It is closer to a public target: launch rockets more cheaply, make radiators lighter, and have the market catch up by then. His strength lies less in getting dates right than in moving money and people across many companies in one direction. If computing moves into space as that order asks, what do we gain, and what bills will come due?
On Earth, electricity is a bigger problem than empty land. A large AI data center needs hundreds of megawatts of power, transmission lines, fiber connections, cooling, and permits, all at once. US data centers used about 4.4% of the country's electricity in 2023. Many forecasts put that share above 10% by 2028.
Building a power plant does not send electricity straight to the servers. Transmission lines and substations need expanding, and grid connections need approval. Power generation and storage projects that recently entered commercial operation in the US took a median of about five years from application to operation. That is not a direct measure of how long data centers wait for a connection, but it shows how fast the grid moves. In that time, AI chips go through two generations. The chip clock runs in quarters and years. The grid clock runs in public hearings and civil engineering works.
The solutions we can use now are on Earth. We can build data centers next to nuclear or gas-fired power plants, or give them power plants of their own. Space is not an answer that lets us skip these options. It is a fairly expensive detour that trades waiting time on Earth for the cost of a ride on a rocket.
Still, space offers something hard to get on Earth. In certain orbits, solar panels can receive sunlight for long stretches, with less interference from weather and night. Computing happens right beside the solar panels, so there is no need for long transmission lines between the power plant and the servers. Costs entered as electricity bills on Earth become launch mass and launch counts in space. The rocket company is, in effect, rewriting the power company's cost sheet.
There is also the advantage of processing data generated in orbit right where it is collected. Instead of sending entire satellite images, weather observations, and space telescope data down to Earth, a facility could pick out just the locations of wildfires, the paths of typhoons, or the movements of ships. Early space data centers are more likely to resemble refineries that lighten the load of data than facilities running chatbots on Earth.
The problem is heat. In a vacuum, there is no air or water to carry it away. In orbit, broad radiators have to emit it as infrared radiation. Even assuming one billion watts of waste heat radiated from a single-sided surface at about 300 K with an emissivity of 1, the required radiating area is about 2.2 km². Solar panels with 30% efficiency and no incidence or wiring losses would need a similar area. Picture a space data center, and it looks less like a building packed with servers than a structure with chips suspended between wings that catch sunlight and black wings that shed heat.
Failures and replacements are an even bigger headache. Radiation and micrometeoroids damage equipment. Even if it keeps running perfectly, older hardware becomes a money loser when newer chips do more computing on the same power. Unless robots can swap modules and retrieve broken equipment, orbital servers will become useless before they break. In a data center on Earth, someone can carry a part into a workroom. In orbit, even replacing a single screw comes with rocket and robot schedules. Running servers starts to look like logistics and shipbuilding.
Cost and communications remain issues too. Project Suncatcher's economic case rests on launch costs falling below $200 per kilogram by the mid-2030s. That is not a price confirmed today. It is an assumption that must hold for the business to work. Even if launch costs fall, constantly uploading vast amounts of training data produced on Earth and downloading the results will clog the communications network. The queue that escaped the power grid forms again at the launchpad and the network.
Suppose these challenges are solved. The story then moves beyond where the servers sit. If one company operates the launch vehicles, satellite communications network, power equipment, and servers, a rocket company becomes a space cloud. What customers buy is no longer a server or a kilowatt-hour of electricity. It is the right to receive a computing result by a set time. In moments when a few minutes change the value of a result, as with weather, elections, or war, the same calculation commands a different price. Tomorrow's computing time could be reserved and traded like airline seats or electricity.
Where the cloud settles changes too. Today's data centers cluster along power plants and undersea cables. In space, the advantage goes to launch sites with frequent flights, orbits with long stretches of sunlight, and places where ground stations are visible at the same time. Just as cities grew around railway stations and ports, the computing industry could gather where launch and communications meet. Which orbits and ground stations you can access becomes as important as national borders.
Good orbital planes, frequencies, communications windows, launch schedules, and maintenance missions are all limited. Companies that establish themselves first will not stop at renting out servers. They will decide who gets to compute, and when. Just as an operating system decides the order in which programs run, an orbital operator will decide the order in which the world's questions are processed. The entity handing out queue numbers changes from a power company to a space infrastructure company.
Imagine a massive typhoon, a military conflict, and a financial crisis all happening on the same day. Typhoon track calculations, reconnaissance image analysis, market risk models, and AI services used by hundreds of millions of people join the same queue. The operator does not even need to seize anyone's servers. Simply pushing a task back a few minutes can delay rescue, defense, and market decisions. Delays in computing become delays in the real world.
At that point, the most expensive technology will not necessarily be the fastest chip. Robots that find failed chips and replace whole modules, ultralight radiators that fold and unfold, and optical communications networks that move large volumes of data between satellites could determine who wins and loses in the industry. Policy will need more than launch permits too. We will need to decide which calculations run first during disasters, who audits the records when an operator changes the order, and who cleans up damaged facilities and pays compensation. The first constitution of space computing may look more like rules for managing a queue than rules for storing data.
Nations will try to secure ‘orbital computing reserves’ just as they stockpile foreign currency and oil. In exchange for granting launch site access and frequency registrations, they may demand computing capacity they can call on immediately in a crisis. Priority access to computing may join weapons and intelligence among the assets allies share. Which side a country is on will show less in treaty language than in whose calculations ran first on the night of a crisis.
For now, these institutions exist only in the imagination. But if computing moves into space, the way we ask about sovereignty will change. Where data was stored will matter less than who got to calculate reality first when it mattered. On satellites in the night sky, typhoon paths, rescue orders, and decisions about war and markets will wait their turn. Who should have the authority to decide that order?
