NASA's Gemma 3-Powered Spacecraft Finds Targets Without Human Help

The Satellite That Thinks for Itself

For a time satellites have been taking pictures of the Earth and sending them back to us. Then people have to look at these pictures to figure out what is going on. This can take a time sometimes days.

Something new just happened. A satellite called Yam-9 made by Loft Orbital did something that no other satellite has done before. It looked at the pictures it took. Found what it was looking for all by itself. It did this because it was asked to using language that people can understand.

The software that made this possible is called NAVI-Orbital. It was made by NASAs Jet Propulsion Laboratory. The brain of this software is Google DeepMinds Gemma 3. This is a kind of computer program that can look at pictures and understand what they mean.

What Makes This Different

Usually satellites send back a lot of pictures and data to Earth. Then people have to look at all of this to find what is important. This is slow and expensive. Sometimes satellites take pictures that're not useful because they are blocked by clouds or are just not interesting.

Yam-9 does things differently. It was asked to look at pictures where the natural environment meets development. It was also asked to find infrastructure around railway hubs. The satellite did this all by itself in space.

Lofts head of AI Paul Lasserre said that this is a deal. "This means we can have satellites that are always watching and can tell us when something is suspicious " he said.

The Hardware Behind the Breakthrough

Yam-9 was launched in 2025 as a test for Lofts AI projects. It has a computer chip called an Nvidia Jetson Orin AGX GPU. This chip is very good at doing tasks in space. The satellite is designed to be used by companies like a platform.

Juan Delfa Victoria, from NASA JPLs AI group led the development of NAVI-Orbital. He had to make the software smaller so it could work on the satellite.

Why This Matters

This is a deal for two reasons. First it can make satellites more useful by looking at the pictures they take and only sending back what is important. This can save a lot of time and money.

Second it shows that we can use AI in space to do tasks. This can help us build more complex AI systems in space in the future.

The Race to Build AI-Powered Constellations

Loft has 12 satellites in space now. They want to build a lot more up to 100 so they can watch the Earth all the time.

They are not the ones doing this. Other companies, like Planet Labs and Kepler Communications are also working on using AI in space.

The Future: Digital Assistants for Astronauts

The idea for NAVI-Orbital started with a researcher named Taran Cyriac John. He was thinking about how astronauts could use assistants when they are exploring the Moon or Mars.

"We want to make it so astronauts can talk to a computer and get help " said Delfa Victoria. "Like in movies, where the computer can understand what you are saying and help you."

For now we know that a satellite can look at pictures and find what it is looking for by itself. This is the start of an era in space exploration.

Summary Table: All Fifteen Stories

# Company/Product Core Story Key Number

1 Alphabet $80B equity raise Berkshire invests $10B $180-190B capex 2026

2 Mach Industries 22-year- founder, defense drones $1.8B valuation

3 Microsoft Build MAI models, Soltera, quantum 1,000x more reliable quantum

4 Mistral AI European sovereignty, 2-year warning 1 gigawatt by 2029

5 Nvidia + Custom Silicon RTX Spark AI PC + in-house chips $725B total capex

6 OpenAI ChatGPT superapp. Codex + IPO prep 900M users

7 Amazon AI Merch AI prompt to custom T-shirts via Alexa Free to use

8 Google AI Plus Price cut $7.99 → $4.99 400GB storage

9 Meta + Reliance 168MW AI data center in Jamnagar 8 GW by 2030

10 Zoho Nathu La Made-in-India server, 5 patents 2,000 servers by 2026

11 Opendoor India shutdown 250 jobs lost AI-native teams 2.36M GCC jobs at risk

12 Anthropic CEO has 1 direct report sister runs ops $1T valuation

13 SpaceX Largest IPO ever $75B raise $1.77T valuation

14 Europe AI Risk 5% global compute vs US 80% €200B plan vs $400B US spend

15 NASA/Loft Orbital First VLM satellite finds targets 50-100 satellites, for real-time coverage