First AI message received from space: Google successfully runs unique Gemma 4 model on a satellite..

Can a satellite in space run an AI model and send the response back to Earth? This is no longer just a theoretical question—Google has confirmed that it is possible. Google has successfully run an AI model on one of its satellites in space and received the complete response on Earth. This was made possible through Google’s ‘Project Suncatcher,’ an initiative exploring the potential for establishing AI computing infrastructure in space.

‘Hello World’: A Major Announcement from Sundar Pichai
Google has successfully executed an AI model on a prototype satellite orbiting Earth, receiving the full response on the ground.
Google CEO Sundar Pichai shared this development on the social media platform X, describing the achievement as a ‘Hello World’ milestone for Project Suncatcher. According to Pichai, the team sent a command to the satellite, activated its onboard Tensor Processing Unit (TPU), ran a prompt through an AI model, and successfully received the complete response on Earth. The response came in the form of a short poem—a haiku.

What is Google’s Project Suncatcher?
Project Suncatcher is a research initiative by the company aimed at determining whether large-scale AI computing infrastructure can be developed in space. Specifically, Google is investigating whether AI models can be operated in space using satellites, and whether multiple satellites could be interconnected to handle substantial AI workloads.
The company’s objective is to assess the reliability of the chips and other systems required for AI computing in a space environment. Google’s Project Suncatcher

Creating a Haiku about Space with the Gemma 4 AI Model
A key component of this Google experiment involved the Gemma 4 open model, which was hosted onboard a satellite. During a communication session with a ground station, the team sent a prompt to this AI model.
The prompt asked the model to compose a haiku—a short, structured poem—about space. The satellite generated a response, which was received by the team as the satellite passed over Svalbard, Norway.
Subsequently, Google tested the system with further prompts, including requests to write limericks (short, rhyming poems) about space and to provide technical status updates on the satellite’s systems.
In this way, the company not only had Gemma compose poetry but also tested the ability to retrieve information regarding the satellite’s internal systems.
Google’s Project Suncatcher

What is Google’s next objective?
Thermal Stress: Investigating how chips perform under the extreme temperature conditions found in space.
Radiation Exposure: Analyzing the impact of space radiation on chips and their computing capabilities.
These tests will help determine whether TPU (Tensor Processing Unit) chips can reliably perform AI-related tasks in space over extended periods.
Preparing to Link Two Satellites by 2027

Google states that if the current tests prove successful, the company may advance to the next major phase in 2027—a milestone that involves linking two satellites together.
Through Project Suncatcher, the company is also exploring the feasibility of creating a network—or "constellation"—of multiple satellites. Such a network could potentially be used to handle large-scale AI workloads in the future.
At present, this remains a research and testing initiative. The company must first demonstrate that chips and other computing systems can operate consistently and reliably in space.
Furthermore, Google believes that satellites in Low Earth Orbit (LEO) can receive near-constant sunlight. According to the company, this would allow these satellites to generate up to eight times more solar power than is possible on Earth.
This is why Google is exploring the possibility of building AI computing infrastructure in space. However, several technical challenges remain to be overcome.

Disclaimer: This content has been sourced and edited from News18 Hindi. While we have made modifications for clarity and presentation, the original content belongs to its respective authors and website. We do not claim ownership of the content.