🔍 Read the full analysis: 5 Takeaways From Google’s First Data Center Satellite Haiku on ThorstenMeyerAI.com
TL;DR
Google says a prototype satellite carrying its data-center-class TPU generated a haiku with its Gemma 4 model during a pass over Svalbard. The test shows that AI inference can run on the hardware in orbit, but the satellite is small, power-limited and far from demonstrating an economical space data center.
Google says a satellite carrying one of its data-center-class TPUs generated a haiku with the Gemma 4 AI model while passing over Svalbard, Norway, marking the first reported operating result from its Project Suncatcher effort to test AI computing in orbit. The demonstration confirms a limited machine-learning workload ran on the prototype; it does not show that space-based data centers are ready to compete with facilities on Earth.
The satellite launched on October 1, 2026, on a SpaceX rideshare mission, according to the source material. Built with satellite company Planet on a standard Planet platform, it carries four TPUs. Google and TechCrunch reported that the payload can use only one chip at a time because of its power and thermal limits, with runs lasting roughly 20 minutes. The satellite produces up to about one kilowatt of power.
After roughly three weeks of commissioning, engineers sent a prompt asking Gemma 4 to write a haiku about space. Google said the satellite sent the three-line response back during a ground-station pass over Svalbard: “Blue marble below / Data streams in silent grace / Starlight feeds my core.” The company also said engineers tested space-themed limericks and asked the system to report its technical status. A Google spokesperson told TechCrunch that the satellite was “operating nominally.”
Google describes the response as the “first-ever AI response from a data center class TPU in orbit.” That wording is narrower than saying it was the first AI system in space: a Starcloud satellite carrying an Nvidia H100 launched in November 2025, and CNBC reported that the company later ran and queried Google’s Gemma model in orbit. Suncatcher’s proposed distinction is a network of closely spaced satellites linked by lasers so they can 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 Demonstration, Not a Space Data Center
The test matters because it moves Project Suncatcher from a research proposal to a hardware demonstration: a Google TPU received a prompt and returned an AI-generated response from orbit. That result gives engineers a platform to study how the chips behave outside Earth’s atmosphere, including under radiation and changing temperatures. It is an early engineering milestone, not evidence that the design can yet support large-scale AI workloads.
The proposed rationale is energy. Google has argued that solar panels in a suitable orbit could produce up to eight times more energy than comparable panels on Earth. The source material cites U.S. Department of Energy estimates that data centers used about 4.4% of U.S. electricity in 2023, with the share projected to reach 6.7% to 12% by 2028. Those figures help explain the interest in alternative power sources, but they do not establish that orbital computing would reduce costs or electricity demand overall.
For companies and the public, the near-term significance is chiefly what the test can reveal about feasibility. A useful orbital computing network would need to move data between satellites and customers on Earth, operate reliably, and justify the cost of launches and replacements. The prototype’s single-chip, short-duration operation leaves those larger questions open.
Popular right now
TCL 75-Inch QM6K QLED Mini LED 4K HDR Smart Google TV, 144Hz
Roborock Q7 M5+ Robot Vacuum and Mop with Self-Emptying Dock
Ninja Espresso and Coffee Maker with Built-In Frother, CFN601
As an affiliate, we earn on qualifying purchases.
From Research Plan to Orbit
Google introduced Project Suncatcher in a research post and preprint on November 4, 2025. The concept called for solar-powered satellites carrying TPUs and linked by optical, or laser, connections. Google’s original next milestone was a learning mission with Planet involving two prototype satellites by early 2027. The first prototype launched earlier than that proposed schedule, but the demonstration remains limited to one satellite.
Google published a peer-reviewed version of its research in the journal Joule on the launch date, according to the source material. The paper describes a possible cluster of 81 satellites operating at about 650 kilometers altitude, with neighboring spacecraft 100 to 200 meters apart. Google said a ground-based test of a pair of optical transceivers reached 1.6 terabits per second; that bench result is not a demonstration of a working, interconnected orbital cluster.
The economics in Google’s research depend heavily on launch prices. The source material says the paper estimates a kilowatt of orbital capacity would cost about $14,700 per year in launch costs at current prices, compared with about $810 if launch costs fell to $200 per kilogram. That lower figure is a modeled scenario, not the prototype’s measured operating cost. The material says reaching the projected launch-cost level would require roughly 1,800 Starship launches.
„“first-ever AI response from a data center class TPU in orbit”“
Costs, Heat and Links Remain Open
The haiku test does not establish how reliably the TPU can run longer or more demanding workloads. Google has said it plans to test performance under thermal stress and radiation exposure, but results from those tests are not included in the source material. The satellite’s power limits and short operating bursts also leave its capacity well below that of a terrestrial data center.
Other barriers include removing heat in a vacuum, servicing or replacing hardware that fails, and maintaining dependable connections between satellites and ground systems. Google’s proposed close formation and laser links have not yet been demonstrated by an interconnected Suncatcher pair in orbit. The source material also cites about 47,000 tracked objects in orbit; it does not specify the tracking date or how that count relates to collision risk for this design.
It is also unclear whether launch costs could fall to the level assumed in Google’s more favorable cost scenario, when that might happen, or how the full cost of operating an orbital network would compare with terrestrial computing. The company has shown that one small payload can run an AI prompt—not that a space-based system is commercially viable.
Thermal Tests and a Twin-Satellite Trial
Google says it will continue testing the prototype, including how its TPUs perform under radiation and thermal stress. The next planned milestone, according to the company, is a pair of interconnected satellites in 2027. TechCrunch reported that the two spacecraft are expected to generate more power and test optical links between satellites.
That trial would address a central part of Suncatcher’s proposal: whether separate spacecraft can exchange data quickly enough to act as a coordinated computing system. Google has not provided results from that future mission, and the source material does not specify a launch date or give a timetable for a larger constellation. For now, the haiku is evidence of a successful early test, while the technical and economic case for orbital AI computing remains under evaluation.
Key Questions
What did Google’s satellite do?
Google says it used a satellite-mounted TPU to run its Gemma 4 model and return a haiku during a pass over Svalbard. Engineers also tested limericks and requested a technical-status report, according to the company.
Was this the first AI system in orbit?
No. The source material says Starcloud launched a satellite carrying an Nvidia H100 in November 2025 and later ran Google’s Gemma model on it. Google’s narrower claim is that this was the first reported AI response from a data-center-class TPU in orbit.
How much computing power does the prototype have?
The satellite carries four TPUs but, according to the source material, can operate only one at a time because of power and heat limits. It produces up to about one kilowatt, with reported runs of roughly 20 minutes.
When will Google test connected satellites?
Google has identified 2027 as the planned milestone for two interconnected satellites. The trial is expected to test optical links, but a specific launch date and results are not yet available in the source material.
Source: ThorstenMeyerAI.com
