SpaceX and Google AI orbit plan
- Google said on May 12 it has discussed future launches with SpaceX for Project Suncatcher, its plan to test orbital AI data centers. - The key date is 2027 — Google says a first prototype launch with Planet Labs is planned before any larger launch deal. - This matters because orbital compute just moved from moonshot theory toward supplier talks, even though economics still favor ground data centers.
Orbital data centers sound like science fiction, but the basic pitch is simple. AI needs huge amounts of electricity, cooling, land, and transmission gear. Those constraints are getting uglier on Earth. So Google and SpaceX are now talking about whether some of that compute could eventually move into orbit. ### What actually happened? On May 12, 2026, Google said it has been in discussions with SpaceX and other launch providers about future launches for Project Suncatcher, its orbital data center effort. The immediate news is not that a deal is signed or satellites are flying. The news is that Google confirmed the talks and tied them to a named internal project. (money.usnews.com) ### What is Project Suncatcher? Project Suncatcher is Google’s research plan for a network of solar-powered satellites carrying TPUs and linked by free-space optical communications. The idea is to build a kind of orbital AI cloud — lots of smaller compute nodes working together instead of one giant machine. Google publicly introduced the project in November 2025 and framed it as a long-range moonshot, not a near-term product. (money.usnews.com) ### Why put compute in space at all? Power is the whole argument. In the right orbit, solar panels can generate energy almost continuously and can be much more productive than on Earth. Space also offers natural radiative cooling and avoids some terrestrial headaches like grid bottlenecks, land fights, water use, and local permitting battles. Basically, the dream is an AI data center that chases sunlight instead of a utility hookup. (research.google) ### So why hasn’t everyone done this already? Because the hard part is not the slogan. It is the economics and the engineering. Google itself highlights radiation effects, orbital dynamics, and the need for inter-satellite links fast enough to behave like a real data center — on the order of tens of terabits per second. And today, launch and satellite build costs still make orbit more expensive than ground-based compute. (research.google) ### Where does SpaceX fit? SpaceX is the obvious launch partner because it already dominates launch cadence and has been pushing its own orbital data center vision. Recent industry reporting says the company has been sharing technical details on a much larger orbital compute constellation, and the project has become part of the broader growth story around SpaceX. That makes Google a very logical customer — or counterparty — even if the two companies are also pursuing overlapping ambitions. (research.google) ### Is there any proof this can work? A little — but only at toy scale compared with a real cloud region. Starcloud launched a single NVIDIA H100 GPU into orbit on a SpaceX rocket in November 2025 and later ran inference workloads using Google’s Gemma model. That does not prove orbital hyperscale is economical. But it does show that “AI in orbit” is no longer just a rendering in a slide deck. (spacenews.com) ### What about the “$900 million” angle? That number points to history, not this deal. Google invested about $900 million in SpaceX back in 2015. Social posts have been using that fact to make today’s talks look like a revival of an old master plan. But the cleaner read is narrower — Google has a real orbital compute project, SpaceX is a plausible launch supplier, and now the companies are talking. (datacenterfrontier.com) ### What should you watch next? Watch for hardware, not hype. The key checkpoint is Google’s planned prototype launch around 2027 with Planet Labs. If that happens and the system can move meaningful data between satellites while surviving radiation and heat constraints, the conversation changes fast. Until then, orbital AI remains a serious experiment with a very expensive bill. (techcrunch.com) ### Bottom line? This is real enough to matter, but early enough to stay skeptical. Google did not announce an orbital AI cloud. It confirmed launch discussions around a moonshot project. That is a step forward — just not the giant leap social media is trying to sell. (money.usnews.com)
