Google says Suncatcher satellite generates its first model responses

Google’s Suncatcher prototype generated its first responses from the onboard Google DeepMind Gemma 4 model this week, the company said in TechCrunch’s October 30 report. Ground engineers asked the model to write a haiku about being in space.

In its 24 September 2026 overview, Google described Suncatcher as long-term research into computing in space, announced the previous year. Google developed the prototype with Planet, an aerospace and satellite-imagery company.

An NPR report carried by KPBS said the prototype launched on 1 October 2026 aboard a SpaceX rocket. Planet was to bring the spacecraft into operation before Google tested its chips.

Commissioning—the checks needed to prepare the satellite for work—tested its subsystems over about 3 weeks. Google said it also asked the spacecraft to report on surrounding orbital conditions, and a spokesperson said the satellite was operating normally.

Power and cooling limit operation

The prototype carries 4 Google Tensor Processing Units, or TPUs, chips designed for machine learning. It generates up to 1 kW of electricity. TechCrunch said the chips’ power demands restrict operation to a single TPU for about 20 minutes at a time, complicating communication with the ground.

The earlier NPR report described a 15-minute operating interval for Gemma and attributed that limit to heat management. Google has not established whether the newer account represents a measured improvement.

Google explains that TPUs concentrate heat in a small area, while the vacuum of space provides no airflow to carry it away. The company is testing heat pipes and radiators to prevent the chips from overheating.

Testing heat and radiation effects

The ongoing test campaign will examine chip performance under thermal stress and actual space radiation, extending Google’s particle-accelerator tests on Earth. Radiation can cause “bit flips,” changes to stored computer values that can corrupt data without triggering typical monitoring tools.

Google’s ground tests suggested radiation would have minimal impact on inference—running a trained model to produce an answer—but pose more difficulty for training a model. The orbital tests will examine those effects in space.

Google says the prototype will help identify what works and where failures occur, so engineers can refine the cooling system and apply the findings to future missions.

A conditional follow-up in 2027

Suncatcher project lead Travis Beals said in the earlier NPR report that the timing of economically viable orbital data centers remained uncertain. Technical feasibility would not make a system practical if it remained too expensive.

If the prototype’s TPUs prove capable, Google proposes launching 2 follow-up satellites in 2027. They would generate more power and demonstrate optical links—laser connections between spacecraft—as a step toward satellite clusters that share computing workloads.