Google switches on an AI chip in orbit and has Gemma 4 write a haiku

What Google actually did

As technews.tw reported, Google said in an official blog post that it has switched on a TPU, its custom AI chip, aboard an experimental satellite. The satellite is part of Project Suncatcher, a long-term research effort to power AI computing in space.

The satellite went up on SpaceX's Transporter-18 launch mission. Over the following weeks Google worked through a full activation checklist. That included powering the host computer and the TPU, then running early machine-learning tests.

A haiku from orbit

During a recent contact with a ground station, Google sent the first prompt to Gemma 4, the AI model on board. It asked for a haiku about being in space. The reply came back as the satellite passed over Norway's Svalbard islands:

"Blue marble below / Data streams in silent grace / Starlight feeds my core."

Google then tried a space-themed limerick and a prompt asking the payload to report its own technical status.

Why this is a cautious test

The poem is a friendly proof of life. The real goal is to see whether the chip survives launch vibration, radiation and swings between hot and cold, and whether the power and cooling systems can keep up.

The satellite has only about 1 kW of continuous power. To avoid straining power and heat management, Google turns the TPU on once every 15 minutes.

Radiation is the big unknown. On the ground, the chip held up well in a particle accelerator. In space, radiation can cause "bit flips", where a value stored in hardware changes. That can corrupt data without setting off ordinary monitoring tools.

What comes next

Google will now run a series of tests on how the TPU behaves under thermal stress and radiation in orbit. If the chip proves it can run workloads steadily in space, that sets up the next goal: linking two satellites together in 2027.