Google, SpaceX discuss orbital data centers

Orbital data centers sound like sci-fi, but this story is now real enough to have named companies, prototype dates, and launch talks. Google confirmed on May 12 that it has been discussing future launches with SpaceX and other providers for Project Suncatcher, its plan to test AI compute in orbit. The big idea is simple to say and hard to do — put solar-powered computing hardware on satellites, then use space itself as the power source and, maybe, part of the cooling solution. ### What is Google actually building? Project Suncatcher is Google Research’s moonshot for machine learning in space. Google laid it out in November 2025 as a network of solar-powered satellites carrying its TPU AI chips, linked together into what is basically an orbital compute cluster. (money.usnews.com) Google’s own post framed this as exploratory research, not a near-term commercial service, but it also put a real milestone on the table: two prototype satellites with Planet, targeted for early 2027. (blog.google) ### What changed this week? The new part is the launch side. Google said it has been in discussions with SpaceX and other companies about future launches for Suncatcher, after a Wall Street Journal report said Google and SpaceX were in talks around the project. That turns a vague moonshot into something more concrete — because orbital compute is impossible without rides to orbit, and SpaceX is the obvious heavyweight there. (blog.google) ### Why would anyone want data centers in orbit? Power is the whole pitch. AI data centers on Earth are running into grid limits, power prices, and cooling headaches. In orbit, satellites can get near-continuous solar energy in the right configuration, and the vacuum of space changes the thermal problem — though not in the magically easy way people imagine. Google’s pitch is that if compute can live where the energy is, some of the bottlenecks that slow terrestrial AI buildouts could ease. (money.usnews.com) ### So is cooling in space actually easier? Not exactly. Space is cold, but empty space does not carry heat away the way air or water does on Earth. Hardware has to dump heat by radiation, which is slower and forces big thermal-management design choices. Then add radiation damage, launch stress, and the fact that chips built for data centers were never meant to live on satellites. Google said its preprint work includes radiation testing for TPUs, which tells you the hard part is not the concept — it’s whether the hardware survives and performs. (blog.google) ### Why is SpaceX such a big deal here? Because launch cost decides whether this is a moonshot or a money pit. SpaceX has the most proven heavy launch cadence, and it is also pursuing its own orbital data-center ambitions. Reuters said this effort is one of the major drivers behind SpaceX’s IPO plans, and last month SpaceNews reported Elon Musk had shared technical details about a future orbital data-center constellation. So these talks are not just vendor chatter — they put two would-be builders of space compute into the same frame. (blog.google) ### Where does Planet fit in? Planet looks like the first test partner, not the final answer to launch. Google’s November announcement tied the early-2027 learning mission to Planet, with two prototype satellites meant to test hardware in orbit. The SpaceX discussions seem to be about future launches beyond that first learning step, or at least alongside other options. That matters because the prototype phase and the scale-up phase are very different businesses. (money.usnews.com) ### What’s the catch? Bandwidth, latency, maintenance, and economics. Training frontier AI models means moving absurd amounts of data, keeping thousands of chips synchronized, and replacing failed hardware fast. Orbit is bad at all three compared with a warehouse next to fiber and a power substation. So the first real use case may be narrower — specialized workloads, edge processing, or proving that some compute can be colocated with abundant solar power. (blog.google) That last part is an inference from the technical constraints and Google’s still-experimental framing. ### Bottom line? This is no longer just Elon Musk free-associating about space. Google has a named program, a prototype target for early 2027, and confirmed launch discussions with SpaceX and others. The hard part is that every advantage of orbital compute comes bundled with brutal engineering penalties. But that’s why the story matters — the AI boom is now pushing companies to test whether the next data-center frontier is not another state, but another altitude. (blog.google)