Read the full analysis: Can A Satellite Write Poetry? Google’s Test And AI In Space on ThorstenMeyerAI.com
TL;DR
Google says its Gemma 4 model generated a short poem aboard a prototype satellite carrying the company’s TPU chip, launched on October 1, 2026. The test shows that a small AI workload can run in orbit, but it does not establish that space-based data centers are practical or economical; power, heat, radiation, communications and launch costs remain major challenges.
Google says its Gemma 4 model generated a three-line poem on a TPU aboard a prototype satellite this week, marking an early test of Project Suncatcher, the company’s effort to study AI computing in orbit. The result confirms a small AI workload ran on Google hardware in space; it does not show that orbital data centers can yet compete with facilities on Earth.
Google said engineers sent a prompt to the satellite during a pass over Svalbard, Norway, then received the response at a ground station: “Blue marble below / Data streams in silent grace / Starlight feeds my core.” The company described the exchange as its “Hello World” moment. Engineers also tested space-themed limericks and asked the payload for technical status, Google said. A Google spokesperson told TechCrunch that the satellite was “operating nominally.”
The spacecraft launched on October 1 aboard SpaceX’s Transporter-18 rideshare mission, alongside more than 100 other spacecraft, according to the source material. Built with satellite company Planet on a standard Planet platform, it carries four TPUs and can generate up to about one kilowatt of power. TechCrunch reported that power and heat constraints mean it can operate one TPU at a time, in short bursts. After roughly three weeks of commissioning, Google said the team powered up the host computer and TPU trays and ran initial machine-learning tests.
The demonstration is limited: generating a short response is not the same as training a large model or operating a network of orbital data centers. Google’s next tests are intended to examine how TPUs handle thermal stress and radiation exposure. The company has described two interconnected satellites in 2027 as a later milestone, if testing proceeds as planned.
Prototype 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 20271 Oct 2026Prototype launchedSpaceX Transporter-18 rideshare, built with Planet; peer-reviewed paper in JouleOct 2026About three weeks of commissioningHost and TPU trays powered up, early machine learning testsWeek of 26 Oct“Hello World”First prompt to Gemma 4; the haiku came down during a pass over Svalbard, NorwayNextHeat and radiation tests in orbitHow the TPUs behave under thermal stress and real space radiation2027Twin 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 GPUTrained a small language model in orbitRan Google’s open model Gemma up there in December 2025Google’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
The Economics Behind Orbital AI
The test matters as a hardware milestone, while the larger proposal is driven by demand for electricity. Google’s research argues that solar panels in a suitable orbit could be up to eight times more productive than on Earth. That is a company research claim about potential solar output, not a demonstration that an orbital computing system can deliver usable power to customers at a lower overall cost.
The scale of the current test illustrates the gap. The satellite’s reported output is about one kilowatt; a nuclear power agreement Google signed on Earth on October 6 adds 890 megawatts, roughly 890,000 times that amount. The source material says US data centers used about 4.4% of national electricity in 2023, while the Department of Energy projects a range of 6.7% to 12% by 2028. Those figures help explain interest in new power sources, but do not establish that moving computation to orbit would reduce electricity demand or costs overall.
Launch economics are a major constraint in Google’s own analysis. Its paper estimates that a kilowatt of orbital capacity would cost about $14,700 a year in launch costs at current prices, compared with about $810 if launch costs fell to $200 per kilogram. The latter would be close to the power cost for US data centers cited in the source material. Google estimates reaching that launch-cost scenario would require roughly 1,800 Starship launches. These are modeled estimates, not current operating costs for a commercial orbital network.
From Research Plan to Orbit
Google introduced Project Suncatcher on November 4, 2025, with a research post and preprint proposing solar-powered satellites carrying TPUs. The concept depends on satellites flying close together and using laser links to share data, so they could act more like a coordinated computing cluster than separate machines. Google’s initial stated milestone was a learning mission with Planet involving two prototypes by early 2027. The October launch moved the first hardware test earlier, but the demonstration remains much smaller than the proposed system.
Google’s TPU is not the first AI chip to reach orbit. The source material says startup Starcloud launched a satellite carrying an Nvidia H100 GPU in November 2025. CNBC reported that Starcloud later trained a small language model and ran Google’s Gemma model in orbit. Google’s specific claim is narrower: it calls this the first AI response from a data-center-class TPU in orbit. Its proposed distinction is the use of close satellite formations and optical links to connect multiple spacecraft for shared computing. Google’s research describes a possible cluster of 81 satellites at about 650 kilometers altitude, but that is a design concept, not the configuration now in orbit.
““the first-ever AI response from a data center class TPU in orbit””
What the Prototype Has Not Proven
The available account does not provide independent technical measurements of the poem-generation test, including its processing time, energy use or detailed operating conditions. Nor does it establish how much work the satellite can sustain beyond short bursts. Google’s own description places thermal and radiation testing among the next steps, so the payload’s performance under prolonged exposure remains unreported.
Other major engineering and business questions remain open: how to remove heat in a vacuum, repair or replace hardware, maintain reliable links to ground stations, and protect systems from radiation. The source material cites about 47,000 tracked objects in orbit, but does not specify a time-stamped estimate or explain how many would pose a direct risk to a proposed Suncatcher constellation. The projected launch costs also depend on substantial future reductions; no commercial operating price or schedule for a full system is established.
Thermal Tests and Satellite Links
Google says it plans a series of tests to measure TPU performance under thermal stress and radiation exposure. If those tests go well, the company’s next stated milestone is a pair of interconnected satellites in 2027. TechCrunch reported that the follow-up spacecraft are expected to generate more power and test optical links between satellites.
Those tests should clarify whether the system can move beyond a single satellite’s brief computing bursts and exchange data between spacecraft. For now, the confirmed result is narrower: a Google TPU in orbit processed a prompt and returned a short poem. Whether that can grow into a reliable, affordable computing network remains unproven.
Key Questions
Did Google’s satellite really write a poem?
Google says it did. The company reported that its Gemma 4 model generated a three-line poem on a TPU aboard the satellite, and that engineers received the response at a ground station.
Was this the first AI system in orbit?
Not according to the source material. Starcloud launched a satellite with an Nvidia H100 GPU in November 2025 and later ran a language model and Google’s Gemma model in orbit, as CNBC reported. Google’s more specific claim is that its prototype produced the first response from a data-center-class TPU in orbit.
How much computing power does the prototype have?
The satellite carries four TPUs and can generate up to about one kilowatt, according to the source material. TechCrunch reported that it can operate one TPU at a time because of power and heat constraints.
When will Google test connected satellites?
Google has identified two interconnected satellites in 2027 as its next milestone, conditional on the current tests. The follow-up is expected to test optical links, but the schedule and technical results are not yet confirmed.
Source: ThorstenMeyerAI.com