🔍 Read the full analysis: The Haiku Google’s Satellite Wrote—and What It Says About AI In Space on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A small satellite built by Google and Planet has run Gemma 4 on a Google TPU in orbit and sent a haiku to Earth. The demonstration shows that the hardware can perform a basic AI task in space; it does not show that orbital data centers are practical or economical. Google’s next tests will examine radiation, heat and links between satellites.
Google says a small satellite carrying one of its data-center-class TPUs has run the Gemma 4 AI model in orbit and sent a haiku to a ground station during a pass over Svalbard, Norway. The October demonstration is the first reported result from Project Suncatcher, Google’s research effort to test whether solar-powered satellites could eventually perform AI computing in space.
Google described the poem as its “Hello World” moment. Engineers uplinked a prompt asking for a haiku about space, and the satellite returned: “Blue marble below / Data streams in silent grace / Starlight feeds my core.” Google said engineers also tested space-themed limericks and asked the system to report its technical status. The poem shows that the payload processed a prompt and returned a response; it is not evidence that the satellite can support large-scale AI workloads.
The prototype, built with satellite company Planet, launched on October 1 aboard a SpaceX rideshare mission. After about three weeks of commissioning, Google said it began early machine-learning tests. TechCrunch reported that the spacecraft carries four TPUs and can generate up to about one kilowatt of power. Power and thermal constraints mean it can operate only one chip at a time, in short bursts. Google said the satellite was operating as expected; a spokesperson described it to TechCrunch as “operating nominally.”
The next tests are aimed at how the chips handle radiation and thermal stress in orbit. Google’s longer-term design involves clusters of satellites connected by optical, or laser, links, so they could share computing work. The current spacecraft is a single prototype, not a connected cluster, and the haiku demonstration did not test that system.
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
Testing AI Beyond Earth
The demonstration matters as a hardware test, not as a new source of computing capacity. It indicates that a Google TPU and an AI model can perform a basic task aboard a satellite and return the result to Earth. That is an early step toward testing whether computing equipment can operate reliably amid space radiation, temperature swings and limited power.
Google’s stated motivation is access to solar energy. Its research argues that solar panels in a suitable orbit could produce more energy than panels on Earth. But power generation is only one part of the cost. Launch, spacecraft construction, communications, cooling and replacement all affect whether orbital computing could compete with terrestrial data centers. The prototype has not established that it can.
For companies and the public, the near-term relevance is the pressure to find enough electricity for growing AI workloads. This mission offers no immediate relief to data-center power demand. Instead, it tests whether a different infrastructure model might be possible in the longer term, with substantial engineering and cost questions still open.
Products Worth Considering
Jackery Explorer 1000 v2 Portable Power Station
Jackery Explorer 300 Portable Power Station, 292Wh
Anker SOLIX S2000 Portable Power Station, 2,010Wh, 1,500W
As an affiliate, we earn on qualifying purchases.
From Research Plan to Prototype
Google introduced Project Suncatcher in a research post and preprint on November 4, 2025. Its initial plan described solar-powered satellites carrying TPUs and linked closely enough to work together. Google then said it aimed to launch two learning-mission prototypes with Planet by early 2027. The single satellite launched in October 2026 is an earlier, smaller demonstration than that planned pair.
Google’s research was also published in the peer-reviewed journal Joule on October 1, according to the supplied source material. The paper discusses possible satellite clusters and estimates their economics under different launch prices. Those figures are projections, not measurements from the haiku mission. The source reports an estimated annual launch cost of about $14,700 per kilowatt at current prices, falling to about $810 if launch costs reach $200 per kilogram. That lower-cost scenario would require a major reduction in launch costs and, in Google’s estimate, roughly 1,800 Starship launches.
The project is not the first attempt to run AI hardware in orbit. CNBC reported that startup Starcloud launched a satellite carrying an Nvidia H100 in November 2025 and later ran and queried Google’s Gemma model there. Google’s specific claim is narrower: its October test was the first AI response from a data-center-class TPU in orbit. Its proposed distinction is a future network of linked satellites, rather than the fact of running an AI model in space.
““Hello World” moment”
— Google, describing the October test
The Economics and Engineering Remain Open
Reliability at scale has not been demonstrated. Google has not yet shown that multiple satellites can coordinate an AI workload over optical links in orbit. The current test also does not establish how long the hardware can operate, how often it would need servicing, or how performance changes under prolonged radiation and temperature exposure.
The cost estimates depend heavily on future launch prices. Google’s lower estimate assumes launch costs fall to $200 per kilogram; that price has not been reached in the evidence supplied here. The source also identifies heat removal in a vacuum, repairs, links between satellites and communication with ground stations as unresolved challenges. It cites about 47,000 tracked objects already in orbit, but does not provide further detail on how that figure affects the project’s risk or operating plans.
Google’s claim about a first TPU response should also be distinguished from a broader claim that AI has never run in space: reports about Starcloud describe earlier orbital use of an Nvidia GPU and the Gemma model. The precise performance, duration and workload achieved in those earlier tests are not fully detailed in the source material.
Radiation Tests and a Two-Satellite Trial
Google says it will run further tests of the TPU payload under thermal stress and radiation exposure. The next stated milestone is a pair of interconnected satellites in 2027. TechCrunch reported that the planned pair is expected to generate more power and test optical links between spacecraft.
Those results should clarify whether the system can move beyond isolated demonstrations toward coordinated computing. Google has not established a launch date for the two-satellite test in the information provided, nor has it shown that the projected cost targets will be met. Until the hardware, communications and economics are tested at larger scale, Project Suncatcher remains a research program rather than a working orbital data center.
Key Questions
What did Google’s satellite do?
It ran Gemma 4 on a Google TPU after receiving a prompt, then sent an AI-generated haiku back to a ground station. Google says engineers also tested limericks and a technical-status prompt.
Was this the first AI model to run in space?
Not according to the source material. CNBC reported that Starcloud launched an Nvidia H100 GPU in 2025 and later ran and queried Gemma in orbit. Google’s claim concerns a response generated by a data-center-class TPU.
Can this satellite power an orbital data center?
No such capability has been demonstrated. The prototype produces up to about one kilowatt and can run one TPU at a time in short bursts. Google’s larger concept depends on future satellites linked together.
Why put AI computing in orbit?
Google is investigating whether satellites with solar panels could provide energy for computing. Whether that could be economical depends on launch costs and unresolved technical challenges, including heat, radiation, repairs and communications.
What is the next milestone?
Google says it will test the current hardware’s response to heat and radiation, then aims for two interconnected satellites in 2027 to test optical links. The source does not provide a specific launch date for that pair.
Source: ThorstenMeyerAI.com
