🔍 Read the full analysis: The AI Behind Google’s Satellite Haiku: What’s Demonstrated And What’s Not 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 its prototype satellite, built with Planet, has run Gemma 4 on a Google TPU and returned a haiku during a pass over Svalbard. The test shows that a data-center-class TPU can run briefly in orbit; it does not demonstrate a networked orbital data center or establish that space-based AI computing is economical.
Google says a small satellite carrying its TPU AI chip has run the Gemma 4 model in orbit and returned a haiku to Earth, a first operational test of the company’s Project Suncatcher concept. The result, reported October 30, demonstrates a brief AI inference task in space—not a working orbital data center or proof that computing in orbit is economical.
Google’s prototype, built with satellite company Planet, launched on October 1 aboard a SpaceX Transporter-18 rideshare mission. After roughly three weeks of commissioning, engineers sent a prompt while the satellite passed over Svalbard, Norway. Google said the satellite returned the three-line poem: “Blue marble below / Data streams in silent grace / Starlight feeds my core.”
The spacecraft carries four TPUs, but TechCrunch reported that its power supply produces up to about one kilowatt, limiting operation to one chip at a time. Google’s test therefore shows a compact system handling short tasks, not the sustained, high-capacity computing associated with a terrestrial data center. Google said the satellite was operating as expected; a spokesperson described it to TechCrunch as “operating nominally.”
Google calls this the first AI response from a data-center-class TPU in orbit. That wording is narrower than saying it was the first AI chip or AI model in space. 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 distinction is its own TPU and its longer-term plan for satellites linked into 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
Testing AI Beyond Earth
The test matters because it moves Project Suncatcher from a research proposal to an initial hardware demonstration. Google is investigating whether solar-powered satellites could eventually supply computing capacity while relying less on electricity and land-based infrastructure. The company’s research argues that solar panels in a suitable orbit could be up to eight times more productive than on Earth.
That possibility is relevant as electricity demand from data centers rises. The source material cites U.S. data centers using about 4.4% of national electricity in 2023 and a Department of Energy estimate of 6.7% to 12% by 2028. Those figures describe the U.S. and are not evidence that orbital computing will reduce that demand. The satellite’s modest power and short operating windows show how far the test is from addressing large-scale workloads.
Cost is a major constraint in Google’s own analysis. Its paper estimates an orbital kilowatt at about $14,700 per year in launch costs at current prices, compared with about $810 if launch costs fall to $200 per kilogram. The latter estimate depends on a future price level, not a demonstrated operating cost. Google’s research estimates reaching that scenario could require roughly 1,800 Starship launches. The haiku confirms a working experiment; it does not settle the economics.
Products Worth Considering
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
LG UltraGear 32GX870A-B 32-inch 4K OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Research Plan to Prototype
Google introduced Project Suncatcher in November 2025 through a research post and preprint. Its concept is a constellation of solar-powered satellites carrying TPUs and flying close enough to exchange data through optical, or laser, links. The aim is for multiple spacecraft to work together, rather than for each satellite to operate only as a standalone computer.
The October launch came before the timetable Google had previously described. In 2025, the company said it expected a learning mission with Planet involving two prototype satellites by early 2027. The current satellite is an earlier, smaller step. Google has published a peer-reviewed version of its research in the journal Joule, but a research paper and a successful single-satellite test do not validate the full constellation concept.
Google’s paper describes possible clusters of 81 satellites at an altitude of about 650 kilometers, flying within roughly one kilometer of one another. It says a bench test of a pair of optical transceivers reached 1.6 terabits per second. That is a ground test of communications equipment, not a demonstration of an 81-satellite cluster operating in orbit. The company’s next proposed step is a pair of interconnected satellites.
““The first-ever AI response from a data center class TPU in orbit.””
The Engineering Gaps Ahead
Google has not yet shown multiple satellites computing together in orbit. It is also unclear how reliably the prototype can run over longer periods, how its components will withstand radiation and temperature extremes, or how heat can be managed in a vacuum. Google says those environmental tests are still to come.
Other open questions include repair and replacement, dependable links between spacecraft and ground stations, and the risk posed by orbital debris. The source material cites about 47,000 tracked objects in orbit; that count is presented without a date or detailed definition, so it should be treated as a reported figure rather than a measure of risk to this particular satellite. The cost estimates also depend on launch prices falling substantially. Neither a commercial deployment schedule nor a cost comparison for a complete operating system has been established.
The “first” claim needs similar precision. Starcloud’s earlier work means Google’s result should not be described as the first AI computation in orbit. Google’s stated claim is about an AI response from a data-center-class TPU. The two efforts also have different designs, so this test alone does not show that Google’s cluster approach will outperform standalone orbital systems.
Thermal Tests and Twin Satellites
Google says it will test how the TPU performs under thermal stress and radiation exposure. The results have not been provided in the source material. If those tests proceed as planned, the next milestone is a pair of interconnected satellites in 2027, according to Google; TechCrunch reported that the pair is expected to generate more power and test optical links between spacecraft.
That follow-up would address a more demanding question than the haiku test: whether separate satellites can exchange data and coordinate computing in orbit. Until such tests are reported, Project Suncatcher remains an early demonstration with major engineering and cost uncertainties. No deployment date or commercial service has been confirmed.
Key Questions
What did Google demonstrate with the satellite?
Google said a satellite carrying a Google TPU ran Gemma 4 and returned a short poem to Earth after a prompt sent during a pass over Svalbard. The reported test demonstrates a small AI task in orbit, not a full data center.
Was this the first AI computation in space?
No. The source material reports that Starcloud launched a satellite with an Nvidia H100 in 2025 and later ran and queried Gemma in orbit. Google’s claim is specifically about an AI response from a data-center-class TPU.
How much computing power does the prototype have?
TechCrunch reported that it carries four TPUs and can generate up to about one kilowatt, but power and heat limits mean it can run one chip at a time. The reported tests have involved short bursts, not continuous large-scale computing.
Does the haiku show that orbital data centers are cost-effective?
No. Google’s research gives different launch-cost scenarios, including about $14,700 per kilowatt per year at current launch prices and about $810 if costs fall to $200 per kilogram. These are estimates, not proof that an operating orbital data center would be competitive.
What is the next milestone?
Google says it plans to test the chip’s response to radiation and thermal stress, followed—if the program proceeds as described—by two interconnected satellites in 2027 to test optical links. The outcome and schedule remain subject to future tests.
Source: ThorstenMeyerAI.com
