🔍 Read the full analysis: A Satellite Wrote A Haiku—What Does That Say About AI In Space? on ThorstenMeyerAI.com
Get hardware and tech essentials 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 prototype satellite carrying its TPU chip ran Gemma 4 and generated a haiku during a pass over Svalbard. 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.
Google says a satellite carrying one of its TPU chips ran the Gemma 4 AI model and returned a three-line haiku to a ground station during a pass over Svalbard, Norway. The October test is the first reported result from Project Suncatcher, Google’s effort to assess whether clusters of solar-powered computing satellites could one day handle AI workloads in orbit.
The satellite launched on October 1 aboard a SpaceX rideshare mission, according to Google’s account. Built with satellite company Planet, it carries four TPUs, but its available power is limited to about 1 kilowatt. TechCrunch reported that the power and heat constraints mean the satellite can operate one chip at a time, in short bursts. Google said the spacecraft had spent about three weeks in commissioning before engineers sent the prompt that produced the poem.
During a ground-station pass, engineers asked Gemma 4 for a haiku about space. The returned lines were: “Blue marble below / Data streams in silent grace / Starlight feeds my core.” Google described the response as its “Hello World” moment. The team also tested space-themed limericks and asked the payload to report its technical status. A Google spokesperson told TechCrunch that the satellite was “operating nominally.”
The demonstration establishes a limited point: an AI model running on Google’s TPU hardware can process prompts in orbit and send results back. It does not show that the satellite can support large-scale AI training, operate continuously, or compete economically with terrestrial data centers. Google’s next stated tests will examine how the TPUs perform under thermal stress and radiation.
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 Test of AI Computing in Orbit
The experiment is a practical test of a proposed way to meet AI’s growing demand for computing power: put some of the hardware in orbit, where solar panels may receive more consistent sunlight than they do on Earth. Google’s research says solar panels in a suitable orbit could be up to eight times more productive than on the ground. That is a theoretical advantage, not a result established by this small satellite test.
The pressure to find additional power sources is real. The U.S. Department of Energy estimates that data centers used about 4.4% of U.S. electricity in 2023 and projects their share could reach 6.7% to 12% by 2028. Those figures describe electricity use on Earth; they do not show that moving computing into space would reduce total costs or energy use. Launch, spacecraft construction, cooling, communications and replacement would all factor into any comparison.
For now, the haiku is best understood as a demonstration of basic operation, rather than evidence of a new data-center business. The larger engineering question is whether many satellites could work together reliably enough to perform demanding workloads. Google’s concept depends on satellites flying close to one another and linking through lasers, rather than each spacecraft operating as a separate computer.
Products Worth Considering
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
LG UltraGear 32GX850A-B 32-inch 4K UHD OLED Gaming Monitor
LG UltraGear 27G640A-B 27-Inch QHD IPS Gaming Monitor, 300Hz, 1ms, G-SYNC Compatible, FreeSync Premium, DisplayHDR 400
As an affiliate, we earn on qualifying purchases.
From Suncatcher to Orbit
Google introduced Project Suncatcher in November 2025, describing a potential constellation of solar-powered satellites equipped with TPUs and linked by optical connections. Its initial target, according to the source material, was a learning mission with Planet involving two prototypes by early 2027. The single satellite launched in October is an earlier, smaller step; it does not yet test a cluster of interconnected spacecraft.
The test also needs a qualification around claims of being “first.” Google calls this the first AI response from a data-center-class TPU in orbit. That is a specific claim about its chip and response, not a claim that AI computing has never taken place in space. 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 orbit.
Google’s proposed architecture differs in its emphasis on linked satellites. Its research sketches clusters of 81 satellites in a roughly one-kilometer radius at an altitude of about 650 kilometers, with neighboring spacecraft separated by 100 to 200 meters. The idea is that laser links could let the satellites exchange data at rates needed for coordinated computing. The paper reports a bench test reaching 1.6 terabits per second with a pair of optical transceivers; that laboratory result is not proof that a large orbital cluster can maintain comparable links in operation.
Cost is another major constraint. Google’s paper estimates that, at current launch prices, launch costs alone would put a kilowatt of orbital capacity at about $14,700 per year. It estimates that the figure could fall to about $810 if launch prices reached $200 per kilogram. The source material says reaching that lower cost would require roughly 1,800 Starship launches. These are modeled estimates and assumptions, not current operating costs for a deployed orbital data center.
“the first-ever AI response from a data center class TPU in orbit”
Limits the Prototype Has Not Tested
The demonstration does not establish how well the chips will perform over long periods in orbit. Google has said further testing will examine radiation exposure and thermal stress, but results from those tests were not included in the source material. It is also unclear how often the prototype can run workloads, how reliably it can communicate with ground stations, or how its performance changes over time.
Several wider obstacles remain unresolved. Equipment in a vacuum has to manage heat without ordinary air cooling; radiation can damage electronics; and repairs are difficult when hardware is in orbit. Google’s envisioned system would also need dependable high-capacity links between satellites and to Earth. The source material cites about 47,000 tracked objects in orbit, but does not specify the data source or date for that count, or explain how that population would affect a future Suncatcher constellation.
The economics are unsettled as well. Google’s cost estimates depend on future launch prices and a large number of launches, while the prototype’s roughly one-kilowatt power output is far below that of a commercial data center. No timeline or demonstrated cost for a functioning, large-scale orbital system was provided.
Testing the Next Satellite Pair
Google says it will continue testing the prototype, including the effects of thermal conditions and radiation on its TPUs. The next milestone described by the company is a planned pair of interconnected satellites in 2027. TechCrunch reported that the pair is expected to generate more power and test optical links between spacecraft.
That follow-up would address a question the haiku demonstration leaves open: whether separate satellites can communicate fast enough and reliably enough to operate as one computing system. The company has not established in the material provided that the pair has launched or that the tests will lead to a larger constellation. For now, Project Suncatcher remains an early research effort, and the first response is evidence of a small in-orbit test—not a working orbital data center.
Key Questions
What did the satellite do?
Google says the satellite ran Gemma 4 on a TPU and returned a haiku after engineers sent it a prompt during a pass over Svalbard.
Was this the first AI system in orbit?
Not according to the source material. Starcloud launched a satellite with an Nvidia H100 in 2025 and later ran and queried Gemma in orbit, as CNBC reported. Google’s “first” claim refers specifically to an AI response from a data-center-class TPU in orbit.
Can this satellite run a full data center?
No such capability has been demonstrated. The prototype produces up to about 1 kilowatt and can run one TPU at a time in short bursts, according to reporting cited in the source material.
What is Google testing next?
Google says it will test the effects of thermal stress and radiation on the TPUs. Its next stated milestone is a planned pair of interconnected satellites in 2027, intended to test optical links.
Does the test prove orbital AI is cheaper or better?
No. Google’s research offers modeled cost estimates that depend on future launch prices, and the single-satellite demonstration does not establish the economics or performance of a large orbital system.
Source: ThorstenMeyerAI.com
