🔍 Read the full analysis: A Poetry Test In Orbit: Google’s Satellite And The Limits Of AI on ThorstenMeyerAI.com
Get audio and creator gear 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
Google says a satellite carrying its TPU chip generated a poem using its Gemma 4 model during a pass over Svalbard. The test demonstrates a small-scale AI workload in orbit, not a functioning space data center: the prototype can power one chip at a time, and cost, heat, radiation, repair and communications remain major challenges.
Google says a satellite carrying one of its data-center-class TPU chips has generated its first reported AI response in orbit: a short poem produced by the Gemma 4 model during a ground-station pass over Svalbard, Norway. The test is an early demonstration for Project Suncatcher, Google’s research effort into solar-powered AI computing in space, but the prototype is far smaller than a data center and does not establish that orbital computing is commercially practical.
Google’s prototype launched on October 1, 2026, on SpaceX’s Transporter-18 rideshare mission, according to the source report. It was built with satellite company Planet on a standard Planet platform. Google said the spacecraft established contact and was operating as expected after reaching orbit. Engineers then spent about three weeks commissioning its systems, including powering up the host computer and TPU trays and running initial machine-learning tests.
For the first publicized prompt, engineers asked Gemma 4 to write a haiku about space. Google said the satellite returned the response as it passed over Svalbard. The lines were: “Blue marble below / Data streams in silent grace / Starlight feeds my core.” The team also tried space-themed limericks and asked the system to report technical status. A Google spokesperson told TechCrunch the satellite was “operating nominally.”
The spacecraft carries four TPUs, but its power supply produces up to about 1 kilowatt, according to TechCrunch. Power and heat constraints mean it can operate one chip at a time, in bursts. Google described the poem as a “Hello World” moment; the company’s stated next tests will examine how the chips 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 Poem Is Not a Space Data Center
The test matters because it moves Project Suncatcher from a research proposal to a working, if limited, in-orbit experiment. Google is testing whether solar-powered satellites could eventually handle AI computing, potentially reducing reliance on electricity and land-based infrastructure as demand for data centers grows. The poem confirms that a TPU-and-model setup can complete a simple task in orbit; it does not show that a large AI workload can be run economically or continuously there.
Scale is the central gap. The prototype’s roughly 1-kilowatt power ceiling is tiny beside the 890-megawatt nuclear power agreement Google signed on Earth on October 6, as described in the source material. Those figures illustrate the difference between a technology demonstration and the power demands of industrial computing. Space might provide strong solar exposure in suitable orbits, but that advantage must be weighed against launch, hardware, communications and maintenance costs.
The project also tests an approach that could differ from running AI on a single spacecraft. Google’s research envisions satellites linked by lasers so they can work together as a cluster. If that architecture proves workable, it could support larger computing tasks than one satellite can handle. For now, that remains a research goal, not a demonstrated capability of this prototype.
Products Worth Considering
Jackery Explorer 2000 v2 Portable Power Station, 2042Wh
Anker SOLIX C300 Portable Power Station, 288Wh
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
As an affiliate, we earn on qualifying purchases.
From Suncatcher Proposal to Orbit
Google first described Project Suncatcher in a research post and preprint paper on November 4, 2025. The proposal centered on constellations of solar-powered satellites carrying TPUs and connected by laser links. At the time, Google said its next milestone was a learning mission with Planet involving two prototype satellites by early 2027. The single-satellite test launched earlier than that stated milestone, but it is not the planned two-spacecraft demonstration.
Google published a peer-reviewed version of its research in the journal Joule on October 1, the launch date reported in the source material. Its paper sketches clusters of 81 satellites within a radius of about one kilometer, at an altitude of roughly 650 kilometers. The proposed satellites would fly close enough for high-capacity laser links. Google’s research reports a bench test reaching 1.6 terabits per second between a pair of optical transceivers; that is a ground test, not a result from the current satellite cluster.
The effort is not the first attempt to run AI hardware in orbit. CNBC reported that startup Starcloud launched a satellite with an Nvidia H100 GPU in November 2025 and later ran and queried Google’s Gemma model in space. Google’s specific distinction is its claim of the first AI response from a data-center-class TPU in orbit. The broader idea of orbital AI computing is being explored by more than one company.
““the first-ever AI response from a data center class TPU in orbit””
Costs and Orbital Risks Remain
Whether orbital AI can compete with terrestrial data centers remains unresolved. Google’s paper estimates that, at current launch prices, putting one kilowatt of computing capacity in orbit would cost about $14,700 a year in launch costs. In a scenario where launch costs fall to $200 per kilogram, the paper estimates about $810 per kilowatt per year, which Google compares with data-center electricity costs. The lower figure depends on a future launch price, not the current cost. The paper’s scenario would require roughly 1,800 Starship launches, according to the source material.
There are also engineering and operational questions. Heat must be managed in vacuum; electronics face radiation; and a satellite cannot be repaired in the same way as equipment on Earth. The system would need reliable links to ground stations and, under Google’s proposed design, between satellites. The source material also cites about 47,000 tracked objects already in orbit, a concern for operating large constellations. How these issues affect cost, uptime and safety has not been settled by a short test from one satellite.
The source material does not provide independent performance measurements, detailed power-use results from the poem test, or a full account of how long the system can sustain workloads over repeated passes. Google’s demonstration is a company-reported milestone, and the prototype’s results should not be treated as proof of the proposed constellation’s performance.
Two-Satellite Test Targeted for 2027
Google says it will next test the prototype’s TPU performance under thermal stress and radiation exposure. If those trials go well, the company’s next stated milestone is two interconnected satellites in 2027. TechCrunch reported that the pair is expected to generate more power and test optical links between spacecraft—steps needed to assess whether separate satellites can operate as a coordinated computing system.
Those tests should provide a clearer measure of whether the concept can move beyond simple, intermittent tasks. The available information does not specify a launch date for the paired satellites or set out a commercial deployment schedule. For now, Google has shown that a small AI system can answer a prompt from orbit; the next milestone is to test the links, power and conditions required for more demanding work.
Key Questions
What did Google’s satellite do?
Google says its satellite used the Gemma 4 model, running on a Google TPU, to answer a prompt with a short poem during a pass over Svalbard.
Is this the first AI system to run in space?
No. The source material says Starcloud launched an Nvidia H100 GPU in 2025 and later ran Google’s Gemma model in orbit. Google’s narrower claim is that this was the first reported AI response from a data-center-class TPU in orbit.
Can the satellite run a full data center?
No. This is a small prototype with four TPUs and up to about 1 kilowatt of power; it can run one chip at a time in bursts. Google’s much larger orbital-computing design remains a proposal to be tested.
What are the biggest obstacles to space-based AI?
Google’s research faces questions about launch costs, heat management, radiation, repairs, ground communications and satellite-to-satellite links. The current test does not resolve those issues or establish commercial viability.
What is Google planning next?
Google says it will test the TPU under thermal stress and radiation exposure. The next stated milestone is a pair of interconnected satellites in 2027, intended to test optical links and operate with more power.
Source: ThorstenMeyerAI.com
