Google in talks with SpaceX

Google and SpaceX are talking about something that sounds like science fiction but is, turns out, a real infrastructure discussion: putting pieces of AI compute in orbit. The immediate news is not that Google has signed a giant deal or is moving Google Cloud off Earth. It’s that Google is exploring whether SpaceX could launch early hardware for Project Suncatcher, Google’s space-based AI infrastructure effort first announced in November 2025. The reason this even comes up is simple — AI demand is running into very physical limits on land, power, cooling, and grid access. ### What is Google actually discussing? The reporting points to talks between Google and SpaceX about launching experimental orbital data-center hardware tied to Project Suncatcher. Google has already described Suncatcher as a research moonshot aimed at scaling machine learning in space with modular, interconnected satellites rather than one giant station. The current talks look like launch logistics for prototypes, not a finished commercial service. ### What is Project Suncatcher? Suncatcher is Google’s idea for running some machine-learning workloads on solar-powered satellites in low Earth orbit. The design Google published leans on many smaller satellites linked together with optical inter-satellite connections, basically treating orbit like a distributed compute fabric. Google has said the early plan includes prototype satellites to test how machine-learning models and TPU hardware behave in space. (bloomberg.com) ### Why would anyone want data centers in space? Power is the whole pitch. In orbit, solar panels can get near-continuous sunlight without weather, nighttime, or the same land constraints that slow terrestrial buildouts. That makes space sound like an escape hatch from the ugly bottlenecks now hitting AI infrastructure — utility interconnection queues, local permitting fights, water use, and the sheer time it takes to build giant campuses on Earth. (research.google) Google has been working those Earthbound constraints too, including demand-response deals and new clean-power-linked data-center plans, which tells you this is a supplement, not a replacement. ### Why SpaceX? Because if you want to put heavy, power-hungry hardware into orbit, launch cost and cadence matter more than almost anything else. SpaceX is the obvious candidate because it dominates commercial launch volume, and it has its own orbital-compute ambitions. Earlier this year, SpaceX sought approval for a huge satellite network meant to support space-based data centers, with filings framing the system as a response to exploding AI-driven data demand. (blog.google) ### Isn’t that weird if they might compete? Yes — but infrastructure markets do this all the time. A company can be your supplier and your rival in different layers of the stack. The interesting wrinkle here is that Google is not betting everything on one rocket company. The current reporting says Google has also explored other launch providers, which suggests it is keeping leverage and treating launch as a procurement problem, not a strategic marriage. (bloomberg.com) ### What are the hard parts? Almost everything after launch. Chips have to survive radiation. Hardware has to reject heat in vacuum, which is much harder than blowing air across racks on Earth. Networking has to work across moving satellites. Repairs are miserable. And latency is not magically solved just because the servers are in space — if users are on Earth, data still has to come down. Google’s own research frames this as an engineering exploration, which is the right level of humility. (bloomberg.com) ### Is this about replacing normal data centers? No. Google is still spending heavily on normal ones, including a $40 billion Texas investment through 2027 and new AI fabric inside its terrestrial cloud. That’s the clearest tell that orbital compute is a long-shot extension of the AI buildout, not the main event. Think of it less like “the cloud is leaving Earth” and more like Google testing whether one painful bottleneck can be moved somewhere with endless sunlight. (research.google) ### Bottom line The real story is not that Google has cracked space computing. It’s that AI demand has become extreme enough that one of the world’s biggest cloud companies is seriously testing whether orbit belongs on the infrastructure roadmap at all. (bloomberg.com) (blog.google)