🔍 Read the full analysis: AI Beyond Earth: What Google’s Satellite Haiku Can Actually Tell Us on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
A satellite built by Google and Planet generated a haiku using Google’s Gemma 4 model on an orbital TPU, demonstrating a small-scale AI computing test in space. The result does not establish that orbital data centers are practical: the prototype has limited power, and costs, cooling, radiation, repairs and satellite networking remain major challenges.
A Google-Planet satellite in orbit has used Google’s Gemma 4 AI model running on a TPU chip to generate a haiku, marking the first reported result from Project Suncatcher, Google’s research effort into space-based AI computing. The demonstration shows that the hardware can run a small AI task in orbit; it does not show that a commercial or large-scale orbital data center is practical.
Google said engineers sent the satellite a prompt during a ground-station pass as it traveled over Svalbard, Norway. The prompt asked Gemma 4 to write a haiku about space. The satellite returned: “Blue marble below / Data streams in silent grace / Starlight feeds my core.” Google described the exchange as its “Hello World” moment for the project. The company also said engineers tested space-themed limericks and requested technical-status information from the payload.
The prototype launched on October 1 aboard a SpaceX rideshare mission, alongside more than 100 other spacecraft, according to the source report. It was built with Planet and is based on one of the company’s satellite platforms. After about three weeks of commissioning, including powering up its onboard computer and TPU hardware, the satellite began early machine-learning tests. Google said it had confirmed contact after launch and that the satellite was operating as expected.
The satellite carries four TPUs but can power only one at a time, according to TechCrunch, which reported the hardware details. Its power generation is up to about one kilowatt. Tests have run in short bursts, described as roughly 15 minutes before launch and about 20 minutes during the reported demonstration. Google said the next tests will examine how the TPUs perform under thermal stress and radiation exposure in orbit.
Infographic 01 · What flew and what it did
One satellite, four chips, one haiku
Google’s Project Suncatcher prototype is the first test of whether its AI data center chips can work in orbit. Here is what is on board and how it got there.
- 4 Nov 2025Project Suncatcher announcedResearch post and preprint; plan: two prototypes with Planet by early 2027
- 1 Oct 2026Prototype launchedSpaceX Transporter-18 rideshare, built with Planet; peer-reviewed paper in Joule
- Oct 2026About three weeks of commissioningHost and TPU trays powered up, early machine learning tests
- Week of 26 Oct“Hello World”First prompt to Gemma 4; the haiku came down during a pass over Svalbard, Norway
- NextHeat and radiation tests in orbitHow the TPUs behave under thermal stress and real space radiation
- 2027Twin interconnected satellitesMore power and laser links between satellites (Google, TechCrunch)
Scale check
The satellite vs. one power deal on Earth
890 MW = 890,000 kW, about 890,000 times the prototype (my arithmetic). The prototype bar is drawn at a visible minimum; to scale it would be invisible.
Not the first AI chip in orbit
Starcloud-1, launched November 2025 (CNBC)
- Carried one Nvidia H100, a data center GPU
- Trained a small language model in orbit
- Ran Google’s open model Gemma up there in December 2025
- Google’s claim is narrower: the first response from a data center class TPU in orbit
Sources: Google, 30 Oct 2026 · Google, 1 Oct 2026 · Google Research, 4 Nov 2025 · TechCrunch, 30 Oct 2026 · TechCrunch, 1 Oct 2026 · Google and Constellation, 6 Oct 2026 · CNBC, 10 Dec 2025.
Powered by Thorsten Meyer AI
A Small Test of Orbital Computing
The demonstration matters because it moves Project Suncatcher from a proposal toward a hardware test: an AI model ran on a Google data-center-class chip in space and returned a response. That is a limited but concrete step toward testing whether solar-powered satellites could perform computing tasks beyond Earth. Google’s longer-term proposal relies on groups of satellites linked by lasers, rather than treating each spacecraft as an isolated computer.
The potential appeal is access to sunlight. Google’s research says solar panels in a suitable orbit could be up to eight times more productive than on Earth. That argument is relevant as electricity demand from data centers grows: the U.S. Department of Energy estimates that data centers used about 4.4% of U.S. electricity in 2023 and could account for 6.7% to 12% by 2028. Those figures describe terrestrial demand, not a demonstrated benefit from moving computing into orbit.
Cost is a major condition. Google’s paper estimates that launch costs at current prices would put a kilowatt of orbital capacity at about $14,700 per year. In a scenario where launch costs fall to $200 per kilogram, the estimate drops to about $810 per kilowatt per year, near the cost of power for U.S. data centers cited in the source. That lower-cost scenario would require a dramatic decline in launch prices; the paper estimates reaching it would involve roughly 1,800 Starship launches. These are research estimates, not a budget or forecast for an operating service.
Products Worth Considering
Logitech MX Keys S Wireless Keyboard, Graphite
Bose QuietComfort Ultra Wireless Noise-Cancelling Headphones with Spatial Audio, Black
Bose QuietComfort Wireless Noise Cancelling Headphones, Sandstone
As an affiliate, we earn on qualifying purchases.
From Suncatcher Proposal to Orbit
Google introduced Project Suncatcher in a research post and preprint paper on November 4, 2025. It described a future constellation of solar-powered satellites carrying TPUs and connected through optical, or laser, links. The intended design would place satellites close enough to exchange data at high rates, allowing them to work together on computing tasks. Google later published a peer-reviewed version of its research in the journal Joule on October 1, 2026, according to the source report.
The first flight was smaller than the constellation concept. Google had previously described a “learning mission” with Planet and a plan to launch two prototype satellites by early 2027. The single-satellite test launched in October 2026, while Google’s next stated milestone is a pair of interconnected satellites in 2027. The company’s research sketches larger clusters, including a scenario with 81 satellites operating within a radius of about one kilometer at an altitude of roughly 650 kilometers. That design remains a proposal, not an operational network.
Google’s result also needs a qualification around claims of being first. The source report says Starcloud launched a satellite carrying an Nvidia H100 GPU in November 2025, and CNBC reported that the startup later ran and queried Google’s Gemma model in orbit. Google’s narrower claim is that this was the first response from a data-center-class TPU in orbit. The distinction is the chip and the proposed interconnected-satellite approach, not the first AI model or AI-capable chip ever operated in space.
““As the satellite passed by Svalbard, Norway, the downlinked poem confirmed success.””
Costs and Orbital Risks Remain
The haiku confirms that the prototype completed a small AI task, but it does not establish how well the system performs across sustained or demanding workloads. The satellite’s limited power means it can operate only one of its four TPUs at a time, and the reported tests have been brief. Google has not yet reported the results of its planned thermal and radiation tests.
Other engineering and operating questions remain open. Equipment in orbit must manage heat without ordinary air cooling, withstand radiation and operate without routine physical repair. A larger system would also depend on reliable links between satellites and between orbit and ground stations. The source report cites about 47,000 tracked objects in orbit, but does not detail how that figure affects collision risk for a Suncatcher constellation. Nor does the demonstration establish that launch prices will reach the level assumed in Google’s lower-cost scenario.
Google’s performance and cost figures are a mix of a reported test, company statements and research estimates. The one-kilowatt capacity and satellite hardware details are attributed to TechCrunch; the eightfold solar comparison and launch-cost scenarios come from Google’s research. No operating cost, commercial timeline or customer service has been demonstrated by this prototype.
Tests Before a Satellite Pair
Google says it will continue testing the current satellite’s TPU hardware under orbital temperature and radiation conditions. The next stated project milestone is a pair of interconnected satellites in 2027. According to the source report, the follow-up spacecraft are expected to generate more power and test optical links between satellites, a key part of the proposed network design.
Those tests should provide evidence about whether separate spacecraft can exchange data at useful rates while operating in orbit. They will not, on their own, resolve the economics of a large constellation or demonstrate the feasibility of routine maintenance. For now, Google’s haiku is a proof that one small AI task ran on an orbital TPU, while the larger idea remains at the research and prototype stage.
Key Questions
What did Google’s satellite do?
It ran Google’s Gemma 4 model on a TPU and returned a haiku after engineers sent it a prompt during a ground-station pass. Google also said engineers tried other short prompts and requested a technical status report.
Was this the first AI system in orbit?
Not according to the source report, which says Starcloud launched an Nvidia H100 GPU in 2025 and later ran Google’s Gemma model in orbit. Google’s more specific claim is the first response from a data-center-class TPU in orbit.
Can the satellite run a full data center?
No such capability has been demonstrated. The prototype produces up to about one kilowatt and can run only one of its four TPUs at a time, in short bursts. Google’s larger satellite-cluster concept remains a research proposal.
Why put AI computing in space?
Google argues that solar panels in a suitable orbit could be more productive than solar panels on Earth, potentially supplying energy for computing. Whether that advantage offsets launch costs and technical challenges has not been established.
What happens next?
Google says it will test how the TPUs handle thermal stress and radiation. Its next stated milestone is a two-satellite mission in 2027 to test higher power generation and optical links between spacecraft.
Source: ThorstenMeyerAI.com
