What happens when you run out of cheap land and power for data centers? Apparently, you start looking at orbit.
Google and SpaceX are in talks to build computing infrastructure in space, according to TechCrunch. The concept would use SpaceX's launch capabilities to deploy server hardware - the physical machines that run AI workloads - into low Earth orbit.
The economics don't favor space right now, and the companies reportedly aren't claiming otherwise. Launching hardware into orbit costs orders of magnitude more than installing it in a ground-based facility. Cooling works completely differently without air. When hardware fails in orbit, you can't send a technician to fix it.
So why have the conversation at all? The pitch centers on a few genuine advantages: orbital infrastructure offers global coverage without running cables under oceans or through countries with hostile regulatory environments. Satellites in low Earth orbit can access continuous solar power. And perhaps most practically, finding sites on the ground that can support hundreds of megawatts of electricity for AI training is getting genuinely difficult - governments are slow-walking data center permits across Europe and parts of Asia, partly over energy grid concerns.
Still Just Talks
Google and SpaceX exploring an idea is not the same as either company committing to build something. Reports at this stage typically represent early feasibility conversations, not signed contracts. The gap between "we are discussing this" and "we are launching servers into space" is very wide, and most of these conversations never become projects.
That said, the fact that a major AI hyperscaler is considering orbital compute at all reflects something real about ground-based supply: demand for AI computing capacity is growing fast enough that expensive, speculative alternatives are worth investigating on paper. GPU cloud pricing has stayed elevated throughout 2025 partly because supply of high-end compute hasn't kept pace with demand from model training and inference - the process of running a model to generate responses.
For anyone using AI tools today, this story won't affect your experience in the near term. The relevant question it raises is whether AI compute supply constraints will translate into sustained high costs for AI services - and based on current trajectories, that pressure isn't easing soon.