Obgist | Nigeria News, Celebrity Scoop, International Gist and Tech News

NASA-backed AI satellite YAM-9 identified Earth targets alone in April, the first vision-language model operating in space.

For the first time, an Earth observation satellite has found what it was looking for all by itself. No ground team. No analysts downloading terabytes of images.

It happened in April, onboard a satellite called YAM-9. And it marks the first confirmed use of a vision-language model, or VLM, in space.

This changes how space cameras work, and how valuable they can be.

How it worked

Usually, satellites take pictures and dump everything back to Earth. Analysts on the ground then sort through the flood, using machine learning tools or just their own eyes.

YAM-9 did it differently. 

The spacecraft was built by space infrastructure firm Loft Orbital. Inside it, a software package called NAVI-Orbital, developed by NASA's Jet Propulsion Laboratory, was running Google DeepMind's Gemma 3.

Gemma 3 is built for edge computing. That means it can run on small hardware, far from any data center. Perfect for orbit.

Researchers simply talked to it in plain English. They asked it to find places where nature meets human development. 

They asked it to spot infrastructure around railway hubs. The model looked at the sensor data right there in space, and classified it correctly.

Why this matters now

There are two big takeaways here.

First, the immediate win. If satellites can triage their own data in orbit, ground teams stop drowning in raw images. You only download what matters. That makes Earth observation faster and cheaper.

Second, the long game. This is proof that real AI can live in space.

"It opens the door to always-on, patrol layers in space," Paul Lasserre, Loft's head of AI, told TechCrunch. 

His pitch is simple: tell a satellite to monitor a border and alert you when something looks suspicious, then have a back-and-forth conversation with it. That is now possible.

Inside YAM-9

Loft Orbital does not build satellites the old way. Its spacecraft are flying platforms that carry payloads for other customers. Think infrastructure-as-a-service, but in orbit.

A recent deal saw Loft build, launch and operate six new satellites for EarthDaily, which will sell the data they collect.

YAM-9 was launched in fall 2025 as Loft's pathfinder for orbital AI. It carries an Nvidia Jetson Orin AGX GPU, one of the top chips being used for compute in space.

The NASA JPL team, led by technical lead Juan Delfa Victoria, built NAVI-Orbital to wrap around the off-the-shelf Gemma 3 model. The big job was slimming it down, cutting libraries and memory use so it could actually run on the satellite.

The race is on

YAM-9 is first, but it will not be alone for long.

Planet Labs already flies satellites with Jetson Orin processors. Right now they are doing simpler object detection, but a spokesperson confirmed they are actively researching VLMs.

Kepler Communications, which runs the largest GPU fleet in space today, would not confirm if it has deployed VLMs, citing partner NDAs. It did note there have been "several undisclosed use cases" on its compute satellites since they launched in January.

Lasserre says Loft is pushing ahead. To get real-time coverage of anywhere on Earth, he estimates a constellation of 50 to 100 satellites like YAM-9. Loft currently operates 12 spacecraft on orbit.

The lessons from running these small models, especially around power and memory management, will shape how bigger AI systems are deployed in space next.

And the vision goes beyond Earth. The NAVI project started when Delfa Victoria and JPL researcher Taran Cyriac John were thinking about astronauts on the moon or Mars.

"You have astronauts with pressurized suits, and they cannot be tapping on a keyboard," Delfa Victoria said. "So how about we provide an assistant, like in video games and in movies, where you see an AI which is interactive?"