Google announces that the Project Suncatcher prototype satellite completed its first TPU inference in orbit
BIBIBI
AT A GLANCE
Google said that the Project Suncatcher satellite used TPU to complete AI inference in space and transmit the results back.
Article
Google sent TPU into space,AI and wrote a poem back from orbit
Beating AI breaking news, Google CEO Pichai announced that October the initially launched Project Suncatcher prototype satellite, for the first time, used data-center-grade TPU to complete AI inference in space and transmit the results back to Earth.
Engineers sent a prompt from the ground, asking the TPU on the satellite to run Gemma 4 and write a haiku about space. As the satellite flew near Norway's Svalbard archipelago, it transmitted three lines of poetry:
*Blue marble below*
*Data streams in silent grace*
*Starlight feeds my core.*
“Blue planet beneath my feet,
Data streams flow silently,
Starlight nourishes my core.”
The satellite carries 4 TPU chips, but due to power limitations, it can currently run only 1 chips at a time, for approximately 20 minutes per run. Although space is cold, there is no air to help dissipate heat from the chips, so they must rely on heat pipes and radiators to radiate heat outward. The effects of cosmic radiation on chip operation also require further testing.
In fact,AI has been to space before. Last year,December the U.S. space-computing startup Starcloud used NVIDIA H100 to run Gemma in orbit and even train a small language model. This time, Google has switched to its own TPU; the next challenges are ensuring long-term stable chip operation and coordinating computation across multiple satellites.
Original link https://m.theblockbeats.info/flash/374769
Key points
01
Google CEO Pichai announced that October the initially launched Project Suncatcher prototype satellite used data-center-grade TPU to complete AI inference in space for the first time and transmit the results back to Earth.
02
Engineers sent a prompt from the ground, asking the TPU on the satellite to run Gemma 4 and write a haiku about space.
03
The satellite carries 4 chips,TPU but due to power limitations, it can currently run only 1 chips at a time, for approximately 20 minutes per run.
04
The satellite relies on heat pipes and radiators to radiate heat outward; the effects of cosmic radiation on chip operation still require further testing.
05
Last year, Starcloud December used NVIDIA H100 to run Gemma in orbit and train a small language model.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
This effort places the AI chips originally intended for data centers on satellites, allowing the satellites to perform inference directly and then transmit the results back to Earth. It remains a prototype validation effort; power supply, thermal management, radiation, and coordination among multiple satellites are the key areas for follow-up.
Why it matters to readers
It demonstrates one possible path for in-orbit computing and also reflects the exploration of Google's proprietary TPU in special environments; however, the available information is insufficient to prove that it has achieved long-term stable operation or commercial viability.
It can be understood as follows: satellites are not only responsible for capturing or transmitting data, but are also attempting to run AI。
Risks and unknowns
No original official Google announcement or independent verification was provided.
TPU Whether it can operate stably over the long term in a cosmic-radiation environment still requires testing.
How multiple satellites will coordinate computation, and whether the related solution is commercially scalable, remain unclear.
Whether the costs of power supply, thermal management, and maintenance are higher than those of ground-based or near-ground solutions remains unclear.
Related Developments
Loading event timeline…
Related concepts
TPU
This term is not in the glossary yet. Browse related concepts in the glossary.
