There are many debates in our society (and right here on DailyKos) about the future of AI and its implications for many things, including the extremely high usage of resources that the data centers used to train and run AIs represent. Many are the stories in recent months about both centers being built and those being rejected by communities worried about their power grids and/or their local water supplies. Even though the current administration, being bought and paid for by the tech billionaires, has no particular interest in doing so, the inevitable rumblings of regulation still are making the AI companies do what the tech bros have always done, seek the greener pastures of times and places where they can escape from regulation and have a free hand to do whatever they want. Their answer to the possible regulation and restriction of the growth of data centers is...go to space.
An article today in the New York Times (gift article) pulls together some of what’s been going on in the latter half of 2025 with AI-involved companies and the idea of putting data centers in orbit. The basic concept is that in orbit there are no restrictions on things like size or power consumption because you’re not having to deal with land area occupied and you’re directly producing power. You essentially sidestep any regulation, permitting, etc. and can build as big as you like. One of the current major players in this field, Starcloud, is pushing this angle very hard in their own promotions. Starcloud is backed by Nvidia, and has actually launched a satellite that was used to train an AI model in orbit. So the feasibility of the plan in a computing sense has been proven as far as that goes. Economically, launch costs are still way too high for it to work at large scale, but if they came down to about $200/kg (as some think they will by the 2030s) it might make sense. The real question is not “can we build functioning data centers in orbit” because that’s an engineering question that is pretty much solved and has been except for this last bit that Starcloud has filled in. The question is, “can we build data centers in orbit that will survive there?”
The basic problem is one that’s been know and talked about for years, but will come into new significance if we start to build large structures (or compound structures, as we’ll see) in particular orbits around the Earth. The problem is that the area around Earth is increasingly filled with debris of various kinds and sizes and, worse yet, the debris is concentrated in the most popular and useful orbits that we use. This was recently brought home in an alarming way when China’s Shenzhou-20 spacecraft was hit by micro-debris (smaller than 1mm) and had a viewing window crack. They had to evacuate the crew rather than risk re-entry with them aboard. But 1mm debris is not the main problem, though it is currently impossible for us to track debris that small or smaller and even debris up to 10cm is still only partially tracked. There are currently around 130 million objects in the 1mm-1cm range, all of which can cause equipment damage and failure at orbital speed for anything unshielded (as solar panels generally are, for instance.) There are at least 1 million objects in the 1cm-10cm range, and though at least some of these are tracked, a complete inventory still does not exist and they can cause serious damage to anything they hit. Most satellites and spacecraft we currently use have some protection from this sort of thing, like protective containers for electronics and ablative coatings on outer surfaces to absorb impacts.
While playing orbital roulette with these objects has not yet caused a catastrophic accident, that may partly be because current satellites and spacecraft are relatively small as well as having some protection. But that’s where we get to the problem with these proposed data centers in orbit. They are not small and will have a much greater problem with orbital debris than current spacecraft. Starcloud’s current models are very small indeed but also of limited power. However, their proposals for future expansion (and the real business model they’re selling, along with Google’s Project Suncatcher) is for extremely large, modular platforms or arrays of platforms in heliosynchronous low Earth orbit. That’s a pretty populous place in Earth’s neighborhood, with a lot of satellite activity (including a lot of military and spy satellites whose exact tracking data is, shall we say, not as generally available as other data.) It’s also a place where debris is more common for exactly the same reason. But putting objects or arrays of objects into that orbit that are kilometers across, as both these companies plans propose, is increasing the odds of a hit by many times. Can large structures like this survive without being constantly hit by debris and losing parts of their equipment? The toll on the huge solar arrays at a minimum would be constant, since you can’t really shield them effectively on the functional side. Then there are the radiators that are needed to disperse heat into space, also necessarily exposed. Even if they could keep the actual electronics safe with traditional methods, that still means these things are exposed to a lot of damage accumulating over time, assuming no catastrophic hits from larger debris. And remember that every hit on one of these thing wouldn’t just be damage to it, it would also be a new event generating even more debris.
But the AI world is already looking at this as their possible solution to Earthly limitations. Musk has expressed interest in it, Google has their test project, and Starcloud is actually making the first moves with backing from the 800lb gorilla of Nvidia. It’s their Galt Gulch for AI, a lawless place where they can do what they want with their technology. That there might be risks that they haven’t considered, or that might make this critical orbit less and less usable for all the other things we need it for probably isn’t on their minds.
