🔍 Read the full analysis: Google Put AI On A Satellite. Here’s What The Haiku Does And Doesn’t Show on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
A Google TPU aboard a satellite built with Planet ran Gemma 4 and returned a haiku after the spacecraft passed over Svalbard, Google said. The test shows a small AI workload can run in orbit, but it does not demonstrate a practical space data center: power, heat, radiation, repair and launch costs remain major challenges.
Google says its Gemma 4 AI model generated and sent back a haiku from a satellite carrying a Google TPU chip, an early test of the company’s Project Suncatcher effort to study AI computing in orbit. The small spacecraft, built with satellite company Planet, returned the response during a pass over Svalbard, Norway, showing that an AI workload could run in space; it did not demonstrate a large or economically viable orbital data center.
Google’s October 30 account described the exchange as the satellite’s “Hello World” moment. Engineers sent a prompt asking Gemma 4 for a space-themed haiku, then received the poem through a ground station. Google also said engineers tested limericks and asked the payload to report technical status. The company characterized the satellite as “operating nominally.”
The prototype launched on October 1 aboard a SpaceX rideshare mission and spent about three weeks in commissioning, according to the supplied reporting. TechCrunch reported that the Planet-based satellite carries four TPUs and can generate up to about 1 kilowatt of power. Power and heat constraints mean it can run one chip at a time, in short bursts; TechCrunch described sessions of roughly 20 minutes during the latest test.
Google’s stated next technical work is to test how TPUs handle thermal stress and radiation exposure in orbit. The company has described a subsequent step involving two interconnected satellites, planned for 2027, to test links between spacecraft. Those tests would be a larger experiment than the single-satellite demonstration, but still would not establish that orbit-based computing can compete with data centers on Earth.
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 AI
The result matters as a hardware demonstration, not as evidence that data centers are about to move into space. It shows Google has powered a data-center-class TPU in orbit, run an AI model, and returned its output to Earth. That tests a basic part of the concept: whether computing hardware can operate in a satellite and communicate results to a ground station.
Google’s broader proposal is motivated partly by the energy demands of AI. Its research argues that solar panels in a suitable orbit could produce more energy than comparable panels on Earth. But the prototype’s roughly 1-kilowatt capacity is tiny beside the power required by a commercial data center. The supplied source compares it with an Earth-based nuclear power agreement adding 890 megawatts; those figures illustrate the difference in scale, not a direct comparison of equivalent systems.
The practical case depends on more than sunlight. Spacecraft must manage heat without ordinary air cooling, withstand radiation, transmit data, and operate without routine repairs. Launching and replacing equipment also carries costs. Until these problems are addressed at a larger scale, the haiku is best understood as a proof of basic operation rather than proof of commercial readiness or lower computing costs.
Popular right now
Shark Clean & Empty Cordless Stick Vacuum with Auto-Empty System, BU3521
Bose QuietComfort Wireless Noise Cancelling Headphones, Sandstone
De'Longhi Rivelia Automatic Espresso Machine with Milk Frother
As an affiliate, we earn on qualifying purchases.
From Suncatcher Plan to Prototype
Google introduced Project Suncatcher in a research post and preprint on November 4, 2025. The concept calls for solar-powered satellites carrying TPUs and linked closely together with laser communications, so they could operate as a computing cluster. Google initially described a learning mission with Planet and two prototype satellites targeted for early 2027. The first prototype reached orbit earlier, on October 1, 2026.
The company is not the first to put AI-capable hardware in orbit. The source material says startup Starcloud launched a satellite with an Nvidia H100 GPU in November 2025; CNBC later reported that the company 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. Its proposed distinction is the plan to link satellites into a cluster, rather than relying on computing aboard an individual spacecraft.
Google’s research paper sketches a possible group of 81 satellites flying within about one kilometer of one another at an altitude of roughly 650 kilometers. That is a proposed system, not what the current test deployed. The source also reports that a bench test of an optical link reached 1.6 terabits per second between a pair of transceivers; that laboratory result should not be confused with a completed satellite-to-satellite network in orbit.
““the first-ever AI response from a data center class TPU in orbit””
Costs and Orbital Limits Remain
The prototype has not answered whether an orbital computing system can deliver useful computing capacity at a competitive cost. Google’s paper estimates that, at current launch prices, a kilowatt of capacity in orbit would cost about $14,700 per year in launch costs; under an assumed launch price of $200 per kilogram, the estimate falls to about $810 per year. Those are research estimates based on different launch-cost assumptions, not measured operating costs from the prototype. The source says the lower-cost scenario would require roughly 1,800 Starship launches.
There is also no reported outcome yet for the planned radiation and thermal tests, or proof that a multi-satellite cluster can sustain the high-speed links needed to share computing workloads. The source identifies repair, communications with Earth, and orbital debris as additional challenges, including about 47,000 tracked objects in orbit. The satellite’s long-term reliability, useful workload capacity, and replacement costs are not established in the available material.
The haiku itself confirms a successful prompt-and-response demonstration as described by Google, but the source does not provide independent telemetry or detailed performance data for the run. It remains unclear how the brief test compares with the same workload on Earth, and how much energy the satellite used to produce and transmit its answer.
Two-Satellite Tests Target 2027
Google says the immediate next phase is to test the prototype under orbital heat and radiation conditions. If that work proceeds as planned, the next milestone is a pair of interconnected satellites in 2027. The source says the two spacecraft are expected to generate more power and test optical links, which would address a key difference between Google’s cluster concept and a single satellite running its own workload.
Those tests should offer a clearer view of whether separate spacecraft can exchange data reliably while computing in orbit. They will not, by themselves, settle the economics of a large constellation. For now, the confirmed development is a small-scale demonstration; the timetable, results of the environmental tests, and feasibility of a larger system remain subject to future testing.
Key Questions
What did Google’s satellite do?
Google says the satellite ran its Gemma 4 model on a Google TPU, answered a prompt with a haiku, and sent the response to a ground station during a pass over Svalbard.
Was this the first AI in space?
No. The supplied reporting says Starcloud launched a satellite with an Nvidia H100 GPU in 2025 and later ran and queried Gemma in orbit. Google’s claim is narrower: a response from a data-center-class TPU in orbit.
Does the test prove space data centers are viable?
No. It demonstrates a small AI prompt-and-response test. It does not establish that a large orbital data center would be affordable, reliable, or able to match Earth-based facilities.
What is Google’s next planned test?
Google says it will test the TPU’s response to thermal stress and radiation. The next stated milestone is a pair of linked satellites in 2027, subject to the company’s plans and test results.
Source: ThorstenMeyerAI.com
