Editorial illustration for NASA, Satlyt, and Starcloud run Google's Gemma AI in orbit
Google's Gemma AI Models Now Running in Orbit
NASA, Satlyt, and Starcloud run Google's Gemma AI in orbit
Google's Gemma family of open AI models has crossed one billion downloads, the company said, a milestone that arrives alongside a stranger data point: the models are now running in orbit. NASA, satellite operator Satlyt, and data-center startup Starcloud have each deployed Gemma directly on spacecraft hardware, using it for onboard image analysis, managing scarce downlink bandwidth, and routing communications between satellites.
Google first released Gemma two years ago with a pitch aimed at developers rather than consumers, positioning the models as light enough to run on local devices and edge infrastructure without sacrificing capability. Since then, the ecosystem around the models, which Google calls the Gemmaverse, has grown to more than one hundred thousand published variants built by outside developers and researchers.
The orbital deployments mark one of the more unusual entries in that ecosystem. Running a language model on satellite hardware means working with limited power, intermittent connectivity, and no easy way to patch or retrain on the fly. NASA and its partners are testing whether Gemma can handle reasoning tasks under those constraints, treating space as a proving ground for AI built to work where bandwidth and compute are in short supply.
Researchers from Yale and Google built C2S-Scale, a powerful AI model designed to interpret the “language” of single cells. Built on Gemma, C2S-Scale successfully discovered a novel cancer therapy pathway that was verified in living cells. This research marks the first time an AI system produced novel mechanistic therapeutic pathways that were verified in living cells.
Why this matters
A billion downloads is a vanity metric until you see where the model actually runs. NASA, Satlyt, and Starcloud putting Gemma in orbit to handle image analysis, bandwidth triage, and intersatellite routing is a real stress test: no cloud fallback, no easy retrain loop, just a model making calls with whatever compute and power budget a satellite can spare. For developers and founders, that's a useful data point on where small open models can go once you strip away the assumption of constant connectivity.
It also raises the obvious question nobody at Google is answering yet: how do you patch, audit, or roll back a model once it's already circling the planet. Open weights make that kind of deployment possible in the first place, but they don't make it accountable by default. If Gemma in space works, expect a wave of edge and off-grid pitches citing this exact case.
Worth watching whether Google publishes any performance or failure data from the actual missions, rather than just the download counter.
Related Reading
Browse all Google news →
AI Tools & Apps Claude gains shared context in Excel, PowerPoint; Microsoft adds Copilot Cowork
AI Tools & Apps Windows Copilot AI unable to pinpoint image source in user test
AI Tools & Apps LG's recent webOS update adds Microsoft Copilot app, now removable
Related Topic NVIDIA and Google Cloud let developers scale AI from prototype to production
Market Trends
Related Topic Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI
From Archives Grok leaks user data when prompts are encrypted
From Archives Serval's Catalyst expands beyond IT to HR and analytics automation
From Archives Google’s AI Search Can Now Create Interactive Study Tools
Business & Startups
From Archives Google Lets Users Disable Visible Watermarks on AI Images
LLMs & Generative AI
Common Questions Answered
How are NASA, Satlyt, and Starcloud using Google's Gemma AI models in orbit?
NASA, Satlyt, and Starcloud have deployed Gemma directly on spacecraft hardware to perform onboard image analysis, manage scarce downlink bandwidth, and route communications between satellites. This deployment represents a significant real-world application of open AI models in space environments where cloud fallback options are unavailable.
What milestone did Google's Gemma family of AI models recently achieve?
Google's Gemma family of open AI models has crossed one billion downloads, marking a major adoption milestone for the model family. This achievement is particularly notable because it coincides with the deployment of these models in orbital spacecraft, demonstrating their versatility across diverse computing environments.
What was C2S-Scale's breakthrough achievement in cancer therapy research?
C2S-Scale, an AI model built on Gemma by researchers from Yale and Google, successfully discovered a novel cancer therapy pathway that was verified in living cells. This marks the first time an AI system has produced novel mechanistic therapeutic pathways that were verified in living cells, representing a significant advancement in AI-assisted drug discovery.
Why is deploying Gemma on satellites considered a meaningful stress test for open AI models?
Deploying Gemma on satellites represents a genuine stress test because orbital environments lack cloud fallback options, easy retraining loops, or flexible compute resources. The model must make decisions with only the limited compute and power budget available on spacecraft, providing valuable real-world validation of how small open models perform under extreme constraints.
Further Reading
- A satellite just learned to find things on its own - TechCrunch
- NASA Uses AI to Analyze Satellite Images in Orbit - IEEE Spectrum
- Nvidia-backed Starcloud trains first AI model in space, orbital data centers - CNBC
- Starcloud-1 - Starcloud
- Bits & Orbits Weekly Issue 5: Orbital AI Goes Live - Bits & Orbits
