A geospatial AI model is running aboard the International Space Station for the first time — and it was researchers from Adelaide who put it there. NASA’s Prithvi model, built in collaboration with IBM, is now processing Earth observation data in orbit, marking a genuine breakthrough in edge AI computing.
The deployment comes from a partnership between Adelaide University and SmartSat CRC (the Australian cooperative research centre for space technology). They’ve taken an open-source model designed for terrestrial data centres and made it work in the most hostile computing environment most of us can imagine.
What Prithvi does in orbit
Prithvi is a geospatial foundation model — think of it as a general-purpose AI that understands satellite imagery, land use patterns, and environmental change. It was already impressive on the ground. But in orbit, it solves a problem that’s been limiting Earth observation for decades: bandwidth.
Right now, satellites collect enormous amounts of data and beam most of it back to ground stations. That’s slow, expensive, and means critical insights arrive hours or days after the event. Prithvi running on the ISS can:
- Analyse data in real-time — detect changes as they happen, not after downlink
- Filter before transmission — send only relevant observations, not raw data floods
- Process across spectral bands — combine visible, infrared, and radar data on-orbit
What is Prithvi? Prithvi is NASA and IBM’s open-source geospatial foundation model, trained on multi-spectral satellite data. It can classify land use, detect environmental changes, and analyse Earth observation imagery across multiple wavelengths. Running it in orbit means processing happens where the data is collected, not after transmission to Earth.
The ANZAC connection
This isn’t just a US achievement with Australian names bolted on. Adelaide University and SmartSat CRC led the deployment engineering — the hard part of making a model work in space. Radiation-hardened computing, power constraints, thermal management, and limited memory all make orbital AI a fundamentally different challenge from terrestrial deployment.
SmartSat CRC is one of Australia’s most significant space technology investments, and this deployment validates their focus on on-orbit processing as a strategic capability. For NZ’s growing space sector — Rocket Lab, Dawn Aerospace, and the emerging satellite services market — this is directly relevant technology.
Why edge AI in space matters
The Prithvi deployment is a proof of concept, but the implications scale fast:
| Capability | Traditional (ground processing) | On-orbit AI |
|---|---|---|
| Latency | Hours to days | Near real-time |
| Bandwidth cost | Full data downlink | Filtered insights only |
| Disaster response | Delayed | Immediate detection |
| Privacy | Data crosses borders | Processed locally |
That last point matters more than you’d think. Many countries restrict who can receive satellite imagery of their territory. On-orbit processing means insights can be generated without the raw data ever leaving the satellite — a significant sovereignty advantage.
For disaster response, bushfire detection, flood monitoring, and agricultural management across the Pacific, real-time on-orbit analysis could be transformative. NZ’s geographic isolation makes satellite coverage critical — anything that reduces latency in that pipeline has immediate practical value.
The bigger picture
Prithvi in orbit is part of a shift from “collect everything, analyse later” to “think where you see.” It’s the same logic driving edge computing on phones and IoT devices — just 400km up. And it’s open-source, which means any nation with satellite capability can adapt it.
The efficiency trend matters here too. Models like Baidu’s ERNIE 5.1 (frontier performance at 6% of training cost) and DeepMind’s AlphaEvolve (0.7% compute recovery) point toward AI that does more with less. In space, where every watt and every byte costs orders of magnitude more than on the ground, that efficiency is the difference between impossible and practical.
🔍 THE BOTTOM LINE
The first AI model running in orbit isn’t a US-only milestone — it’s an ANZAC engineering achievement that points toward a future where satellites think for themselves. For NZ, the implications are direct: faster disaster detection, better agricultural monitoring, and a potential niche in on-orbit processing for the Pacific region. The question isn’t whether edge AI goes to space — it’s whether NZ catches that wave or watches from the shore.
❓ Frequently Asked Questions
Q: What does this mean for NZ? NZ’s space sector is small but growing fast. On-orbit AI processing could be a genuine niche — Rocket Lab launches, NZ-based ground stations, and Pacific-focused satellite services. The Prithvi model being open-source means NZ organisations can adapt it for regional needs without starting from scratch.
Q: Can this model run on smaller satellites? The ISS has relatively powerful computing hardware. Scaling to smaller CubeSats requires model compression — but the efficiency trends (smaller models, better performance per watt) are moving in the right direction. Expect this within 2-3 years.
Q: Is this the same Prithvi model from 2023? Yes — Prithvi was first released by NASA and IBM in 2023 as an open-source geospatial foundation model. The breakthrough here is deploying it in orbit, not the model itself. The Adelaide/SmartSat team’s contribution was making it work in the space environment.
