🔍 Read the full analysis: What We Can Learn From Google’s Satellite-Generated Haiku on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Google says a satellite carrying its TPU chip generated a haiku with the Gemma 4 AI model during a pass over Svalbard in late October 2026. The test shows that a small AI workload can run in orbit, but it does not establish that space-based computing is practical or economical at scale.
Google says a satellite carrying one of its data-center-class TPU chips generated a haiku with the Gemma 4 AI model and sent it to a ground station during a pass over Svalbard, Norway, in late October. The test is an early result from Project Suncatcher, Google’s effort to study whether groups of solar-powered satellites could eventually perform AI computing in orbit.
The spacecraft launched on October 1 on a SpaceX rideshare mission, according to the source account, and was built with satellite company Planet. After several weeks of commissioning checks, engineers sent the model a prompt asking for a haiku about being in space. Google described the returned poem as its “Hello World” moment for the project. The company also says engineers tested space-themed limericks and asked the payload to report its technical status.
The satellite carries four TPUs, but its reported power supply of up to about one kilowatt means it can run only one chip at a time. TechCrunch reported that the system operates in bursts, with tests lasting roughly 20 minutes this week. Google said the spacecraft was “operating as expected”; a Google spokesperson told TechCrunch it was “operating nominally.” These statements describe the prototype’s reported status, not a full-scale orbital data center.
Google’s research plan centers on connecting nearby satellites with optical, or laser, links so multiple machines could work together. The current test, however, was a small demonstration: a prompt was processed aboard one satellite and the answer was sent down to Earth. It does not show that an orbital cluster can carry out large-scale AI training or compete with a terrestrial data center.
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
What the Haiku Test Establishes
The test moves Google’s proposal from research toward hardware in orbit: it indicates that a satellite can operate a company’s TPU and return the output of an AI prompt. That is a meaningful engineering milestone for space-based computing, but its scope is narrow. One short response does not test sustained workloads, large-scale model training, or the costs of running a network of satellites.
The appeal Google identifies is energy. Its research argues that solar panels in suitable orbits could produce more power than comparable panels on Earth, while demand from data centers is growing. The source material cites U.S. data centers as using about 4.4% of national electricity in 2023 and a Department of Energy projection of 6.7% to 12% by 2028. Those figures explain the interest in alternatives; they do not show that moving computing into orbit would relieve electricity demand soon.
For readers, the distinction is between an experiment and a service. The haiku demonstrates that a small AI task can be completed in orbit. It says little about whether satellite computing could become reliable, affordable infrastructure, or whether it would change the availability or cost of AI products.
Popular right now
KitchenAid Classic Series 4.5-Quart Tilt-Head Stand Mixer, Onyx Black
Dyson Airwrap i.d. Multi-Styler for Straight and Wavy Hair
Shark Rotator Lift-Away Vacuum with DuoClean PowerFins (LA555) – Dark Grey
As an affiliate, we earn on qualifying purchases.
From Research to Orbit
Google introduced Project Suncatcher in a research post and preprint paper on November 4, 2025. The concept described solar-powered satellites carrying TPUs, flying close together and exchanging data through laser links. Google then identified a learning mission with Planet as a future step, with two prototypes expected by early 2027, according to the supplied account. The October launch brought one prototype to orbit earlier than that stated milestone, but on a smaller scale than the proposed satellite clusters.
Google is not the first company to put an AI-capable chip in orbit. The source account says startup Starcloud launched a satellite carrying an Nvidia H100 GPU in November 2025; CNBC later reported that it had run and queried Google’s Gemma model in orbit. Google’s stated distinction is narrower: it calls this the first AI response from a data-center-class TPU in orbit. The project’s proposed approach—linking satellites to operate together—is a separate ambition from running an AI task on one spacecraft.
The economics remain a major constraint. In Google’s paper, the estimated annual launch cost for a kilowatt of orbital capacity is about $14,700 at current launch prices, compared with about $810 if launch costs fall to $200 per kilogram. The latter is close to the power cost for U.S. data centers cited in the source material. That lower estimate depends on much cheaper launches; the paper’s scenario would require roughly 1,800 Starship launches, according to the account. These are projections and assumptions, not observed costs for an operating orbital data center.
““As the satellite passed by Svalbard, Norway, the downlinked poem confirmed success.””
— Google, as quoted in its account of the test
Costs and Orbital Limits Remain
The demonstration does not establish how reliably the chip can run over long periods, how much useful work it can complete, or what the full cost would be after launch, communications, maintenance and replacement are included. Google’s next technical tests are intended to examine heat and radiation, but results were not provided in the source material. A satellite in vacuum must also shed heat, while repairs in orbit are difficult and links between satellites and ground stations impose operating constraints.
Google’s research also describes tight satellite formations, but the source does not establish that such a constellation has been deployed or that its proposed data rates will be achieved in space. The paper’s cost figures depend on future launch prices, and the estimate of roughly 1,800 Starship launches is a scenario rather than a completed launch plan. The source cites about 47,000 tracked objects in orbit as another concern; it does not specify how that count was defined or how it changes the project’s risk.
It is also unclear from the supplied reporting how much computing the haiku required, whether the model ran entirely on the satellite, or how the system’s performance compared with an equivalent ground-based task. Those details matter before the test can support claims about efficiency or commercial value.
Thermal Tests and a Two-Satellite Trial
Google says it plans further tests of TPU performance under thermal stress and radiation exposure. The next stated milestone is a pair of interconnected satellites in 2027, according to Google’s account. TechCrunch reported that the two spacecraft are expected to generate more power and test optical links between satellites.
That trial would address a different question from the haiku test: whether separate spacecraft can exchange data fast enough to operate as a coordinated computing system. The company has not yet shown that such a network can sustain larger AI workloads, meet its projected costs, or operate at a scale comparable with data centers on Earth. Until further results are reported, the haiku is best understood as a successful small-scale demonstration, not evidence that orbital AI computing is ready for routine use.
Key Questions
What did Google’s satellite do?
Google says the satellite used a TPU chip to run its Gemma 4 model on a prompt asking for a space-themed haiku, then sent the response to a ground station during a pass over Svalbard.
Was this the first AI system in orbit?
No. The source account says Starcloud launched a satellite with an Nvidia H100 GPU in November 2025 and later ran and queried Google’s Gemma model there. Google’s claim is specifically about an AI response from a data-center-class TPU in orbit.
Can the satellite run all four of its TPUs at once?
Not according to the reported hardware details. The satellite has four TPUs, but its power supply is up to about one kilowatt, so it can run one chip at a time, in short bursts.
Does the test show orbital data centers are cheaper?
No. Google’s paper provides cost estimates that depend heavily on lower future launch prices. The reported haiku test does not establish the total cost, long-term reliability or commercial value of computing in orbit.
What is Google planning next?
Google says it plans more tests of heat and radiation performance, followed by a proposed two-satellite trial in 2027 to test connections between spacecraft. The outcome and timing of that milestone remain subject to future updates.
Source: ThorstenMeyerAI.com
