🔍 Read the full analysis: Google’s Space AI Experiment, Told Through A Haiku on ThorstenMeyerAI.com
Get smart everyday buys 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 its Gemma 4 model generated a haiku aboard a small satellite carrying Google TPUs, marking an early test of Project Suncatcher. The demonstration shows that a data-center-class TPU can run in orbit, but the satellite’s limited power and unresolved costs, heat, radiation and connectivity challenges leave large-scale space computing unproven.
Google says its Gemma 4 AI model generated a haiku aboard a small satellite carrying the company’s tensor processing units, a test that marks an early operational result for Project Suncatcher. The satellite’s response came during a ground-station pass over Svalbard, Norway, and demonstrates a limited AI workload running in orbit—not a functioning space-based data center at commercial scale.
Google described the exchange as the satellite’s “Hello World” moment. Engineers sent a prompt asking Gemma 4 to write about being in space; 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 payload to report its technical status. The company said the spacecraft was operating as expected.
The prototype, built with satellite company Planet, launched October 1 on a SpaceX Transporter-18 rideshare mission. After roughly three weeks of commissioning, Google said it powered up the host computer and TPU trays and ran early machine-learning tests. TechCrunch reported that the satellite carries four TPUs but can operate only one at a time, with runs limited by power and heat. It generates up to about 1 kilowatt.
Google calls the result the first AI response from a data-center-class TPU in orbit. That wording is narrower than saying AI has never run in space: Starcloud launched a satellite carrying an Nvidia H100 in November 2025, and CNBC later reported that the company ran and queried Google’s Gemma model in orbit. Google’s distinction is its use of a TPU and its longer-term plan for satellites to work together as a computing cluster.
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 Computing
The demonstration matters because it moves Google’s space-computing idea from research toward hardware tested in orbit. If solar-powered satellites could eventually host AI workloads, companies might gain another way to supply computing capacity as demand for data-center electricity grows. But this test does not establish that orbital computing is cheaper, more reliable or ready to ease pressure on terrestrial power systems.
The scale difference is substantial. The prototype produces up to about 1 kilowatt, according to TechCrunch. By comparison, Google’s nuclear-power agreement announced October 6 adds 890 megawatts on Earth—about 890,000 times the prototype’s maximum power output. The comparison illustrates the gap between a technology demonstration and energy infrastructure, not a direct measure of how much useful computing each project provides.
Google’s research argues that sunlight can be more productive in certain orbits than on Earth, where weather, night and atmospheric conditions affect solar generation. That potential has to be weighed against the cost of putting equipment into orbit, keeping it cool and connected, and replacing or repairing hardware. For businesses and the public, the relevant question is not whether a chip can answer a prompt from space, but whether a whole system can deliver useful computing at a competitive cost.
Products Worth Considering
Dell 34-Inch S3425DW Curved WQHD USB-C Monitor
Canon SELPHY CP1500 Wireless Compact Photo Printer (Black)
Apple iPad Air (2025) 13‑inch Wi‑Fi 256 GB – Space Gray (M3 chip)
As an affiliate, we earn on qualifying purchases.
From Suncatcher Paper to Satellite
Google introduced Project Suncatcher in a research post and preprint on November 4, 2025. Its proposal is to place solar-powered satellites carrying TPUs in close formations and connect them with optical, or laser, links. The goal is to distribute computing across multiple spacecraft rather than treat each satellite as an isolated machine.
The company had previously described a “learning mission” with Planet involving two prototype satellites, with launch planned by early 2027. The first satellite reached orbit sooner, on October 1, 2026, and Google said it had established contact and the spacecraft was operating as expected. Google also published a peer-reviewed version of its research in the journal Joule that day.
The concept remains distinct from simply placing an AI chip in orbit. Google’s research sketches clusters of 81 satellites flying within a radius of about one kilometer at an altitude of roughly 650 kilometers, with neighboring spacecraft 100 to 200 meters apart. Google says a bench test of a pair of optical transceivers reached 1.6 terabits per second. That laboratory result is not evidence that a complete satellite cluster has achieved the same performance in orbit.
Google’s paper also lays out a cost hurdle. It estimates launch costs for orbital power at about $14,700 per kilowatt per year at current prices, compared with about $810 if launch costs fall to $200 per kilogram. The paper’s lower-cost scenario is a projection, not the price of operating this prototype. The source material says reaching the projected launch economics would require roughly 1,800 Starship launches.
““the first-ever AI response from a data center class TPU in orbit.””
Power, Heat and Cost Questions
The test does not show how well TPUs will perform over extended periods under radiation and thermal stress. Google has said those conditions are the subject of further tests. The prototype’s limited power also restricts operation to one TPU at a time, and the source material describes bursts of roughly 20 minutes during the recent test.
Other engineering and operating questions remain open: how a larger system would reject heat in a vacuum, maintain reliable links to ground stations, and handle failures when technicians cannot readily repair a spacecraft. Orbital debris is another concern; the source material cites about 47,000 tracked objects already in orbit, though it does not specify the source or date for that count.
The cost case is also unresolved. Google’s figures depend on future launch prices, and its lower-cost estimate is conditional on a substantial reduction in the cost of lifting payloads. The first satellite’s successful response does not establish the cost per unit of computing, how much power a full cluster would produce, or whether a satellite network could compete with ground-based data centers.
There is also a limit to what can be concluded from the haiku itself. It shows that a prompt was processed and a response returned, as Google reported. It does not by itself verify the performance, durability or economics of the larger system described in the company’s research.
Thermal Tests and Two Satellites
Google says it will test how the TPUs handle heat and radiation exposure in orbit. The company’s 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. These tests would address important parts of the proposed system, but would still be preliminary relative to operating a larger constellation.
The results will help show whether the chips can keep working under orbital conditions and whether satellite-to-satellite links can support computing across spacecraft. Google has not established in the source material a date for a commercial service or a decision to build a full-scale orbital data center. For now, the haiku is evidence of a small system performing a basic task in orbit; the next milestones are tests of endurance, power and connection.
Key Questions
What did Google’s satellite do?
Google says the satellite ran Gemma 4 on a TPU and returned a haiku after engineers sent it a prompt during a ground-station pass over Svalbard. The company also reported tests involving limericks and technical-status prompts.
Was this the first AI model in orbit?
No. The source material says Starcloud launched a satellite with an Nvidia H100 in November 2025, and CNBC later reported that Starcloud ran and queried Google’s Gemma model in orbit. Google’s claim is specifically about an AI response from a data-center-class TPU in orbit.
How much computing power does the prototype have?
TechCrunch reported that it carries four TPUs and can generate up to about 1 kilowatt, but power and heat limits mean it operates one chip at a time. It is a prototype, not a large orbital data center.
Why is Google testing AI computing in space?
Google’s research explores whether satellites could use solar energy and optical links to form computing clusters. The company argues that solar panels in a suitable orbit may be more productive than on Earth, but the cost and engineering case has not yet been proven at scale.
What happens next?
Google says it will test TPU performance under thermal stress and radiation exposure. Its next stated milestone is two interconnected satellites in 2027, which are expected to test higher power and optical links, according to TechCrunch.
Source: ThorstenMeyerAI.com
