🔍 Read the full analysis: What A Space-Bound Haiku Can Tell Us About AI—and What It Can’t 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 a satellite carrying its data-center-class TPU generated its first response in orbit this week: a haiku produced by its Gemma 4 model. The test is an early milestone for Project Suncatcher, not proof that orbital data centers are practical; power, heat, radiation, connectivity and launch costs remain major constraints.
Google says a satellite carrying its data-center-class TPU has generated and returned its first AI response from orbit: a three-line haiku produced by the Gemma 4 model. The test, conducted during a pass over Svalbard, is an early operational milestone for Project Suncatcher, Google’s research into solar-powered AI computing in space—not evidence that orbit can yet support large-scale data centers.
The prototype, built with satellite company Planet, launched October 1 on a SpaceX rideshare mission. After several weeks of commissioning, engineers sent Gemma 4 a prompt asking for a space-themed haiku. Google said the satellite returned the response during a ground-station pass. The poem was: “Blue marble below / Data streams in silent grace / Starlight feeds my core.” Google described the exchange as its “Hello World” moment.
The satellite carries four TPUs but can produce up to about 1 kilowatt of power, according to hardware details reported by TechCrunch. That limit, along with heat constraints, means it can run only one chip at a time in short bursts. TechCrunch reported bursts of about 20 minutes during the current tests. Google said the satellite was “operating as expected”; a Google spokesperson told TechCrunch it was “operating nominally.”
Google’s next tests are intended to examine how the TPUs perform under thermal stress and radiation exposure. The company’s longer-term concept involves closely spaced satellites using optical, or laser, links to work together as a computing cluster. Google has described a follow-on mission with two interconnected satellites in 2027. That would test a more demanding part of the proposal than running a model on a single satellite.
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 result shows that a commercial AI model and a data-center-class processor can be operated in orbit and communicate with a ground station under test conditions. It does not show that space-based computing can replace or meaningfully supplement Earth-based data centers at scale. The demonstrated workload was a short text response, and the satellite’s power and operating time are limited.
Google’s proposal addresses pressure on electricity supply as demand from AI and other data-center workloads grows. The U.S. Department of Energy estimated that data centers used about 4.4% of U.S. electricity in 2023 and projected a range of 6.7% to 12% by 2028. Those are estimates and projections for electricity use on Earth, not evidence that orbital facilities would reduce that demand. Space-based systems would still need to transmit data, survive harsh conditions and justify the cost of getting equipment into orbit.
The test matters chiefly as a technical checkpoint: it moves Project Suncatcher from a published concept to a working, if small, orbital demonstration. For companies and the public, the key question is not whether a satellite can write a poem, but whether future systems can deliver useful computing at a cost and reliability that compete with terrestrial facilities.
Products Worth Considering
Dell 34-Inch S3425DW Curved WQHD USB-C Monitor
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
Apple MacBook Pro 2023 with M2 Pro Chip, 14‑inch, 16 GB RAM, 512 GB SSD – Space Gray
As an affiliate, we earn on qualifying purchases.
From Research to Orbit
Google introduced Project Suncatcher in November 2025, describing a possible network of solar-powered satellites equipped with TPUs and linked by lasers. Its initial stated milestone was a learning mission with Planet involving two prototype satellites by early 2027. The first prototype launched earlier, on October 1, 2026, but it is a limited test rather than the interconnected system envisioned in the research.
Google’s idea is not the first attempt to put AI computing hardware in orbit. CNBC reported that startup Starcloud launched a satellite carrying an Nvidia H100 GPU in November 2025 and later ran and queried Google’s Gemma model there. Google’s specific claim is that its mission produced the first AI response from a data-center-class TPU in orbit. The distinction is the chip and the proposed architecture, not a claim that no AI model had operated in space before.
Google’s research argues that solar panels in a suitable orbit could be up to eight times more productive than on Earth. But the economics depend heavily on launch prices. The source material cites Google’s estimate that a kilowatt of orbital capacity would cost about $14,700 a year in launch costs at current prices, compared with about $810 if launch costs fell to $200 per kilogram. Google’s paper estimated that reaching the lower-cost scenario would require roughly 1,800 Starship launches. These are modeled estimates, not measured operating costs for a deployed orbital data center.
““the first-ever AI response from a data center class TPU in orbit.””
Scale, Cost and Reliability
The haiku test does not establish how much useful computing the satellite can sustain, how reliably it can operate over time, or whether multiple satellites can coordinate a demanding workload. Google says it plans tests of radiation exposure and thermal stress, but results from those tests were not provided in the source material. The prototype’s power limits also leave open how much work future hardware could perform without major changes.
The cost estimates depend on future launch prices and assumptions about system design. The source material does not establish that launch costs will reach the level used in Google’s lower-cost scenario, or that the overall expense of building, maintaining and replacing orbital systems would match Earth-based power costs. Other unresolved challenges include heat management in vacuum, repair limitations, communications with the ground and the presence of roughly 47,000 tracked objects in orbit, according to the supplied source.
It is also not yet clear how the planned satellite links will perform outside a test setting, or whether they can provide the bandwidth and dependable service needed for AI computing. Google’s concept remains a research program; no deployment schedule or commercial service commitment is described.
Two-Satellite Tests in 2027
Google says it will continue testing the current satellite, including its TPUs’ response to radiation and temperature conditions in orbit. The next stated milestone is a planned pair of interconnected satellites in 2027. TechCrunch reported that the follow-on spacecraft are expected to generate more power and test optical links between satellites.
Those tests will help show whether the project can progress beyond a single satellite running short workloads. Until Google reports results on linked operations, sustained performance and costs, the haiku remains a demonstration of a narrow capability—not proof that large AI data centers can be moved into space.
Key Questions
What did Google’s satellite do?
It ran Google’s Gemma 4 model on a TPU and returned a short, space-themed haiku during a ground-station pass, according to Google.
Was this the first AI model to run in orbit?
No. The supplied reporting says Starcloud launched a satellite with an Nvidia H100 in 2025 and later ran and queried Gemma in orbit. Google’s claim is narrower: its satellite produced the first response from a data-center-class TPU in orbit.
Can this satellite run a full AI data center?
No such capability has been demonstrated. The prototype produces up to about 1 kilowatt and runs one TPU at a time in short bursts, according to details reported by TechCrunch.
Why is Google testing AI computing in space?
Google is researching whether solar-powered satellites could support computing as demand for electricity from data centers grows. Whether orbital systems can provide useful computing at competitive cost remains unproven.
What is the next milestone?
Google has said it plans further tests of the current satellite and a two-satellite mission in 2027 to test interconnected operations. The performance and cost results have not yet been reported.
Source: ThorstenMeyerAI.com
