Imagine a scene from a Friday evening: your takeout dinner has just been placed on the table when a friend sends you a news article in a group chat, the headline simultaneously mentioning Google, AI, and satellites. Before you even unpack your chopsticks, a sci-fi movie automatically starts playing in your mind: a row of glowing machines orbiting the Earth, busy answering what humans want to eat tonight. This image is quite adorable; let it linger in your mind for a moment.
Next, a very reasonable question might appear in the group: "So AI has to go on a business trip too?" Putting the TPU on track is indeed well-suited to the image of a business trip. However, this colleague's luggage, transportation, and work environment all need to be checked item by item. Let's treat this as reading an itinerary for this special business trip: first, see where to go, what to do, and then what report to submit upon return.
When you entrust a problem to AI, is it possible that the devices responsible for the computation could one day be placed in orbit around the Earth? Google's "Solar Catcher" project is exploring this. According to a report published by iThome on September 25, 2026, Google is preparing to launch a prototype satellite equipped with a TPU to bring this concept into the practical testing phase. [S1]
For the average reader, the most noteworthy aspect of this news is what questions the test intends to answer. For a chip to go into space, it must overcome several hurdles, from being launched by a rocket to its continuous computation in orbit. Understanding these hurdles makes it easier to grasp the significance of the subsequent results.
First, let a prototype satellite undergo inspection.
The report indicates that Google announced on September 24th that it expected to launch its first prototype satellite equipped with Google's Tensor Processor (TPU) via SpaceX's Transporter-18 mission "next week." This timeline is based on that report and describes the mission plan announced at that time. [S1]
This mission, developed in collaboration with Planet, primarily aims to collect on-orbit operational data of the TPU. Key observations listed in the report include: the chip's ability to withstand the intense vibrations of a rocket launch, space radiation, and extreme temperatures. [S1]
By breaking these items down, the roles of the tests become clear. Launch vibrations affect whether the equipment can successfully complete its journey to orbit; radiation and temperature introduce issues to the operating environment. When reading test messages, first ask which stage it's referring to, and then see what criteria were used, to avoid letting a simple "test successful" statement encompass everything.
To use a more relatable analogy, a prototype satellite is like sending a complete set of equipment to an on-site acceptance test. The acceptance test checks not only whether the equipment is present, but also whether it functions as intended. This reflects an understanding of the mission's purpose: breaking down the plan into verifiable issues and then answering them one by one with actual records.
Let's start by imagining a pop-up coffee stall.
Imagine a pop-up coffee stall preparing for an out-of-town event. This is just a simple analogy: the machines in the shop are usually working perfectly, but this time they need to be packed up, moved, and reopened in a new location. Before setting off, the machines need to be checked; once they arrive, the necessary equipment needs to be set up; and only after starting business do they realize how well the whole process is going.
At this point, if someone says, "The box has been delivered successfully, congratulations on the event's success," the person in charge of dispensing the cups will probably slowly look up. Arrival is cause for celebration, but a long list of tasks remains: Will the machine work? Will the first cup be made? Will we be able to consistently meet the pre-ordered quantities? Each checkmark has meaning and its own place.
Prototype satellites are also suitable for this segmented perspective. Launch can be seen as a transportation segment in a journey, and the work performed in orbit as the mission segment after arrival. This analogy is merely to help organize the reading order; it doesn't apply the structure of a coffee machine to TPU. Having this mental to-do list in mind while reading the news makes it less likely that seeing the words "send it up" will immediately trigger a complete mental update on the entire project.
Moreover, the act of trying it out on a small scale is visually appealing in itself. Planning the process beforehand is often more effective at clarifying issues than having everyone stand in an open space saying, "It should be okay, right?" Viewing Suncatcher from this perspective, the presence of prototype testing is stronger than any pretty conceptual illustration.
Suncatcher wants to utilize the sunlight in its orbit
According to reports, Project Suncatcher is a long-term research project announced by Google in November 2025. Its concept is to use near-continuous sunlight in low Earth orbit to provide power, and then connect multiple TPU-equipped satellites into a constellation via high-speed laser communication. [S1]
"Constellation" here refers to a system composed of multiple satellites. Looking back from this goal, the plan can be broken down into several interconnected questions: where will the power come from, how will the chips operate, and how will the different satellites exchange data? The solar power, TPU, and laser communication mentioned in the report correspond to these questions, respectively. [S1]
This also explains why it's worthwhile to build a prototype satellite first. When the final concept involves multiple satellites, understanding the performance of the computing equipment helps in discussing how the subsequent systems should be combined. This is an observation derived from the project structure: first understand how each part works, then discuss what the overall service might look like.
Group Mini-Drama: Are there group administrators for zodiac-related groups?
Below is a hypothetical chat between friends, which is suitable for organizing the terms mentioned earlier.
Xia: "Astrology sounds so romantic, so TPU is starting to offer horoscopes too?"
Ah-Cen: "This constellation is a system composed of many satellites. Today's lucky item, let's start with the test records."
Xia: "So the sun is responsible for powering the circuits, the TPU is responsible for the operation, and the laser is responsible for enabling communication?"
Ah-Cen: "We can start by remembering them like this, like putting name tags on each character. But we've made progress on the Rizhao problem, so we still need to see how well the computation and communication are doing."
Xia: "I understand. There's a difference between someone posting a picture in the group and actually submitting the group assignment."
That's enough of the humor. The last sentence is actually quite suitable for reading large-scale technology plans. When something has many stages, it's best to keep each stage's own verb: power supply, execution, connection, verification. The clearer the verbs are, the easier it is for the discussion to focus. You can also directly ask, "What action did this message add?" This way, the next time you see an update, you'll know where to put it in the schedule.
What were the first questions asked during the ground-based proton beam test?
Prior to this orbital test, Google had conducted a ground-based radiation-related test. The report indicates that researchers at the Crocker Nuclear Laboratory at UC Davis used a proton beam to irradiate a TPU running AI workloads and observed the chip's performance. [S1]
Preliminary results show that the Trillium TPU can withstand a total ionizing radiation dose exceeding the dose expected during a five-year space mission. This "five years" is a benchmark used to describe radiation dose; when reading the numbers, focus on "how much dose was received." [S1]
For this study, irradiating the chip while it was performing its task was a noteworthy testing scenario. It placed the radiation conditions and the computational work within the same observation. To compare different results later, one could follow the same line of thought, examining what tasks the chip was given during the test and how the researchers recorded its responses.
Ground testing and in-orbit testing each provide different observation points. The former allows performance under specific conditions to be examined, while the latter sends the equipment to the mission site. Following this logic, what SolarCatcher needs to accumulate next is a record that allows design, testing conditions, and actual operation to be compared with each other.
Fun Facts About Tools: The Number Five Years is Accompanied by an Important Noun
If we were to pick a small piece of knowledge from this news that we could easily add during a chat, it would be "total ionizing radiation dose." The five years mentioned in the report refers to the dose the mission was expected to receive; the Trillium TPU mentioned in the test was exposed to a proton beam while performing AI workloads. [S1] Reading the nouns next to the numbers together makes the meaning much more complete.
Imagine a luggage testing label. If the label says how much weight it can withstand, the reader will naturally look at the kilograms; if it says how many times it can be opened and closed repeatedly, the reader will look at the number of times. Everyone calls it a durability test, but the questions it actually answers are different. Reading chips can also retain this attention to detail: the numbers are eye-catching, but the units and conditions tell people how to use it.
Another highlight is that the tool is placed in a specific context: the TPU is performing its work while being irradiated with a proton beam. [S1] When you read this sentence, you can add a small asterisk next to it to remind yourself to pay attention to what the device is doing at that moment when you look at the test results. This way of reading doesn't require memorizing a whole page of terminology, but it can gradually transform sentences that originally sound like exam questions into a vivid picture of action.
A few minor points that are easy to misunderstand
The first one is that "five years" is automatically translated by the brain as "five-year warranty." When reading quickly, it's really easy to add a guarantee to it. But here, the sentence can be read in its entirety: radiation dose received beyond the expected five-year mission duration. This puts the focus back on the amount being compared in the test, rather than assigning another task to the numbers. [S1]
The second is that "having power" immediately gives people the assurance that "everything can get started." Returning to the pop-up coffee stall analogy, when the venue announces that the power is ready, everyone breathes a sigh of relief; then, they still need to follow the procedures to confirm each piece of equipment. The concept of the solar catcher is also well-suited for segmented reading: first understand the direction of power supply, then examine the chip's operation, then the communication between satellites, and progress on each topic independently.
The third approach is to think of the prototype as a scaled-down version of the full service. One of the most interesting aspects of prototyping is helping people ask more specific questions. When reviewing it, you can appreciate what it has delivered without rushing to plan all its future work. This follow-up approach is actually more engaging because each round of testing connects to the previous one, like watching a TV series where each episode has a new plot.
What can we do next to pursue this plan?
To determine how far the actual measurements have progressed, we can start with specific questions: When did the satellite enter orbit? What tasks did the TPU perform? How long did each task take? How were temperature, workload, and performance recorded together? These observations allow subsequent discussions to focus on what the equipment actually did and facilitate connecting the results from different stages.
Another area worth exploring is inter-satellite communication. According to reports, Google plans to launch two satellites in 2027 to test high-speed laser communication. This plan shifts the focus from single devices to connections between multiple satellites. [S1]
SolarCatcher has placed AI computation within a very specific engineering problem: once the chip is in orbit, how will it function, and how will it collaborate with other satellites? As testing unfolds, the most interesting aspects will be the comparable and traceable records. These will allow "space AI computation" to gradually become a set of problems with practical conditions and results to discuss.
Weekend mini-exercise: Make a satellite business trip note.
I had a little free time on the weekend, so I grabbed a sticky note and did some easy organization. No tools or engineering drawings needed; the back of a crumpled receipt will do. First, write "TPU Track" in the middle, then draw three circles around it and fill in "Departure," "Work," and "Contact" respectively.
Next, add the content from the article. The vibration during launch connects to departure; the chip's operation in orbit connects to operation; high-speed laser communication connects to contact. Solar power supply can be represented by a small sun next to the main image. This is just a categorization to aid memory; there's no need to strive for every circle to be the same size. Drawing them crookedly actually gives it a more journal-like feel.
Next, write a question for each circle. For the departure circle, you could write "Where is the mission progress updated to?" For the work circle, write "What work has been completed, and for how long?" For the communication circle, write "How to verify the coordination between multiple satellites next?" If you only want to make a very short version, you can choose the question that interests you the most. The key is to condense your curiosity into a simple, understandable question.
Finally, let's play a little game: Imagine the next news story only says "Successfully completed an AI task." Try pasting it into the diagram we just created. It will be closer to the "task" circle; the other circles can be left for their respective progress. After organizing it this way, when you encounter long news headlines with many English abbreviations, you'll know better where to start reading. Your brain finally won't have to replay the entire sci-fi movie every time.
If you'd like to find someone to discuss this with, you can ask a friend to summarize the project in one sentence and then say in another sentence which aspect they're most interested in following. It's also great if two people are interested in different aspects; one might want to look at the chips, while the other wants to see how satellites work together. This allows you to discuss the same news from different angles. The most comfortable moments in tech conversations are often when everyone can find a small area that interests them.
Postscript: I'll leave this space for the next update.
The name "Suncatcher" certainly evokes the image of a little sprite chasing sunlight with a net. This image can be preserved as a memorable one, while a more practical note can be attached next to it: Prototype Satellite, Operational Data, Phased Verification. Placing the endearing name alongside concrete questions creates both a vivid picture and a clear direction.
The next time you see news about space AI in a group chat, don't rush to send a bunch of exclamation marks. Take a look at which section of that business trip note has new content, then throw the most interesting sentence back into the group chat—that's enough for conversation. Being able to slowly understand how a big plan is broken down into smaller steps is also a very enjoyable aspect of following tech news.
source
- [S1] Google's Space Data Center project is entering the testing phase, with the launch of its first AI satellite scheduled for next week. | iThome | Published: 2026-09-25.