🔍 Read the full analysis: What Google’s Orbital Haiku Experiment Shows About Space-Based AI 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 small satellite carrying Google TPUs generated a haiku with the Gemma 4 model during a ground-station pass this week, Google said. The test is an early demonstration for Project Suncatcher, not a working orbital data center: power, heat, radiation, connectivity and launch costs remain major hurdles.
Google says a satellite carrying its Tensor Processing Units (TPUs) generated a haiku in orbit this week, a small but concrete test of the company’s Project Suncatcher plan to explore AI computing in space. The response came from Gemma 4, Google’s open AI model, during a pass over Svalbard; the company describes it as the first AI response from a data-center-class TPU in orbit.
The prototype launched on October 1 aboard SpaceX’s Transporter-18 rideshare mission, alongside more than 100 other spacecraft, according to the source material. Google said the satellite established contact and was operating as expected. After roughly three weeks of commissioning, engineers sent a prompt during a ground-station pass and received the poem: “Blue marble below / Data streams in silent grace / Starlight feeds my core.” Google called the exchange its “Hello World” moment.
The satellite was built with Planet and is based on a standard Planet platform. TechCrunch reported that it carries four TPUs and can generate up to about one kilowatt of power. Because of power and thermal constraints, the satellite can operate only one chip at a time, in short bursts. The test shows that an AI model can run on a TPU in orbit; it does not show that a large-scale AI training workload can be performed there.
Google’s next planned step is to test how the TPUs respond to radiation and thermal stress. The company has said it aims to launch two interconnected prototype satellites in 2027 to test links between spacecraft. That would move the project beyond a single satellite’s computing demonstration and toward a key part of Suncatcher’s concept: multiple satellites cooperating as a computing 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
From One Chip to Orbital Clusters
The haiku matters as an engineering milestone, not as evidence that orbit is ready to replace terrestrial data centers. Google is testing whether solar-powered satellites could eventually supply computing capacity as demand for AI and electricity grows. In its research, the company argues that solar panels in suitable orbits could be more productive than panels on Earth, where night, weather and atmospheric conditions affect generation.
The scale gap is substantial. The prototype’s reported output of about one kilowatt is enough for limited tests, not a commercial computing facility. The source material contrasts that with a 890-megawatt nuclear power deal Google signed on Earth in October, but those figures describe very different projects and should not be read as equivalent systems. For now, Suncatcher is testing hardware and operating conditions; the business case depends on whether future launch costs, satellite power and networking can support far larger workloads.
For readers, the development offers a glimpse of one possible response to the energy demands of AI infrastructure. But any effect on electricity use, data-center construction or consumer services remains speculative. The system would need many spacecraft, dependable high-speed links and lower costs before it could materially change where AI computing happens.
Popular right now
Shark Rocket Corded Stick Vacuum (HV301) – Gray/Orange
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
De'Longhi 15-Bar Pump Espresso Machine with Milk Frother & Temperature Control
As an affiliate, we earn on qualifying purchases.
How Project Suncatcher Developed
Google introduced Project Suncatcher in November 2025, describing a potential constellation of solar-powered satellites equipped with TPUs and linked by lasers. The company’s early research envisioned closely spaced spacecraft that could work together, rather than relying only on individual satellites to run separate tasks. Google’s initial public milestone was a learning mission with Planet, then planned for early 2027; the single-satellite test arrived earlier, but at a smaller scale than the proposed constellation.
Google is not the first organization to put an AI-capable chip or model in orbit. The source material says Starcloud launched a satellite carrying an Nvidia H100 GPU in November 2025, and CNBC later reported that it ran and queried Google’s Gemma model in space. Google’s narrower claim is about a response produced on a data-center-class TPU. Its longer-term distinction is the proposed use of tightly clustered satellites connected by optical links, a design that has not yet been demonstrated in orbit.
Google’s published research also sets out a cost challenge. It estimates that, at current launch prices, launch costs alone would amount to about $14,700 per kilowatt per year for orbital capacity. The paper estimates that the figure could fall to about $810 if launch costs reached $200 per kilogram. Those are research estimates, not the operating cost of a functioning orbital data center; the lower figure would require a major change in launch economics.
““the first-ever AI response from a data center class TPU in orbit””
Costs and Reliability Remain Open
The test does not establish how well TPUs will perform over extended periods in orbit. Google says it plans to examine radiation exposure and thermal stress, but the source material does not report results from those tests. It also provides no detailed account of the satellite’s long-term reliability, maintenance strategy or performance on sustained workloads.
The proposed system would depend on links between satellites and communication with ground stations. Google has described optical links as part of its plan, but the two-satellite test intended to examine them has not yet happened. It is also unclear how launch, replacement and servicing costs would compare with the cost of building and powering data centers on Earth. Google’s cost estimates are projections based on specified launch-price assumptions, not a complete commercial accounting.
Orbital debris is another stated concern: the source material cites about 47,000 tracked objects in orbit. That figure is a count, not a measure of the probability of a collision for this satellite. The available information does not specify how Suncatcher would manage debris risks as a constellation grows.
Two-Satellite Test Planned for 2027
Google says it will first run additional checks on the prototype, including tests of TPU behavior under radiation and heat. The next stated milestone is a pair of interconnected satellites in 2027. TechCrunch reported that the pair is expected to generate more power and test optical links between spacecraft, though the source material does not provide a launch date or full mission schedule.
Results from those tests will help show whether the project can progress from a single chip responding to prompts toward coordinated computing across satellites. Until Google reports those results—and clearer figures on performance, reliability and cost—the haiku remains an early demonstration rather than proof that space-based AI data centers are commercially viable.
Key Questions
What did Google’s satellite do?
Google says it ran the Gemma 4 model on a TPU aboard a satellite and returned a haiku during a ground-station pass. The company described the exchange as an early “Hello World” test.
Is this the first AI system in space?
No. The source material says Starcloud launched a satellite with an Nvidia H100 in 2025 and later ran Google’s Gemma model on it. Google’s claim concerns an AI response from a data-center-class TPU in orbit.
Can this satellite run a full AI data center?
No such capability has been demonstrated. TechCrunch reported that the prototype produces up to about one kilowatt and can run one of its four TPUs at a time in short bursts.
What is Google planning next?
Google says it will test the prototype under radiation and thermal stress, then aims to test two interconnected satellites in 2027. The pair is expected to examine optical links between spacecraft.
Is orbital AI computing cheaper than data centers on Earth?
That has not been established. Google’s research estimates launch costs of about $14,700 per kilowatt per year at current prices, with a much lower estimate under a future launch-cost assumption. Those figures do not establish the total cost of a commercial orbital data center.
Source: ThorstenMeyerAI.com
