Google Tests TPUs in Orbit as Microsoft Expands Copilot

Google is testing whether artificial intelligence computing can eventually move beyond traditional data centers and into space. Its Project Suncatcher prototype satellite, developed with Planet Labs, launched aboard SpaceX’s Transporter-18 rideshare mission in October 2026. The satellite carries Google’s Tensor Processing Units (TPUs), specialised chips designed to accelerate machine-learning workloads. Google has confirmed contact with the spacecraft and says it is beginning experiments to study how the hardware performs in orbit.

Meanwhile, Microsoft is giving Copilot a broader role across work, software development and Windows. The company’s redesigned Copilot experience brings together Home, Code and Autopilot, while its Windows plans include AI-assisted access to local files and the ability to perform multi-step tasks. The two developments address different challenges: Google is researching a potential new location for future AI infrastructure, while Microsoft is bringing AI capabilities closer to the applications and devices people already use.

Key Takeaways

• Google’s Project Suncatcher is testing whether specialised AI chips can operate reliably in orbit under radiation, temperature extremes and launch-related stress.
• The prototype satellite carries four TPUs and is designed to run limited AI experiments, including workloads using Google’s Gemma model.
• Google is investigating whether solar-powered orbital computing could eventually support scalable AI infrastructure, but commercial deployment remains a long-term research goal.
• Microsoft’s redesigned Copilot introduces Home, Code and Autopilot to combine assistance, application development and task delegation.
• Microsoft is also working to let Copilot interact more directly with local files and Windows functions, subject to feature availability and rollout.
• Neither initiative represents a finished transformation of AI computing: Google is testing hardware in space, while Microsoft’s new capabilities are being introduced gradually.

Google Project Suncatcher: Why Put AI Chips in Space?

AI computing requires substantial processing power and electricity. Large data centers bring together thousands of accelerators, networking equipment and cooling systems to train and run AI models. As demand for these services grows, technology companies are exploring ways to improve the cost, energy efficiency and scalability of the infrastructure behind them.

Project Suncatcher examines whether some of that computing could eventually be placed in orbit.

Google’s long-term concept involves satellites equipped with AI processors and powered by solar energy. In principle, space offers access to sunlight without the day-to-day constraints of terrestrial weather and the availability of land. A sufficiently capable orbital network could potentially perform certain AI workloads while relying on solar power.

But the idea involves much more than sending a computer into space. The hardware must survive launch, operate in a radiation-rich environment, manage heat without conventional airflow and communicate with other satellites or ground stations. The economics of launching, maintaining and replacing equipment also need to make sense.

Google’s prototype is intended to help answer some of those questions through real-world testing. The company says the mission will collect data about how TPUs withstand the physical stresses of spaceflight and the radiation and thermal extremes encountered in orbit.

The research is a long-term experiment, not evidence that Google is moving its existing data-center operations into space.

How Google’s Orbital TPU Experiment Works

Tensor Processing Units are Google’s custom-designed processors for machine-learning workloads. They are used to accelerate calculations involved in training and running AI models. In Project Suncatcher, the company is testing whether this class of hardware can function in a satellite environment.

Google developed the prototype in partnership with Planet Labs. The spacecraft launched on the Transporter-18 rideshare mission operated by SpaceX, and Google confirmed that it had established contact with the satellite.

Reporting by NPR and KPBS describes the prototype as carrying four TPUs. The satellite is designed to run a version of Google’s Gemma AI model for short periods, providing an opportunity to test actual AI processing in orbit rather than relying exclusively on laboratory simulations.

Radiation and hardware reliability

Space exposes electronics to radiation that can interfere with computing hardware and affect reliability. Earth-based data centers operate in a very different environment, where equipment can be shielded and maintained under controlled conditions.

Google’s experiments can help determine how its chips behave under orbital radiation exposure and whether future designs require additional protection or other changes.

Thermal management

Heat is another major challenge. AI processors generate heat when they operate, and conventional server cooling often relies on moving air across heat sinks and other components. In the vacuum of space, air cannot circulate in the same way.

Orbital systems must instead transfer heat through suitable materials and radiators that release thermal energy. Managing the balance between processing power, cooling capacity and available electrical energy will be essential if space-based AI systems are to scale.

Power and communications

Solar power is central to Google’s long-term vision, but generating electricity is only one part of the equation. Future systems would also need reliable ways to distribute power, coordinate computing tasks and transfer data between spacecraft and users on Earth.

High-bandwidth communications could become especially important if satellites are expected to work together as a distributed computing network. Google has discussed further experiments to investigate these challenges, but the current prototype is an early step rather than a functioning orbital data center.

Can Space-Based AI Data Centers Become Practical?

Project Suncatcher raises the possibility that orbital computing could eventually supplement data centers on Earth. However, its commercial feasibility is far from established.

Solar energy could offer an attractive power source in orbit, but the total cost of an AI system includes launch expenses, satellite construction, radiation protection, communications, cooling, maintenance and replacement. Hardware that fails in space may be considerably harder to repair than equipment inside a terrestrial facility.

Latency is another consideration. Many AI services require rapid communication between users, applications and computing infrastructure. Satellite links introduce additional communication constraints, and the suitability of orbital computing will depend on the type of workload and where its users are located.

Not every AI task needs the same infrastructure. Some applications prioritise low latency, while others can tolerate delays or process information in batches. Space-based systems could eventually make sense for selected workloads, but that would require evidence that their benefits outweigh their additional engineering and operating costs.

Google has described Project Suncatcher as a long-term research effort. The current mission is intended to generate experimental data that can inform future designs, not to demonstrate that orbital computing is already cheaper or more efficient than conventional data centers.

The distinction matters as technology companies increasingly explore alternative ways to meet rising AI demand. A successful prototype would be an important engineering milestone, but it would not automatically establish a commercially viable business model.

Microsoft Expands Copilot With Home, Code and Autopilot

While Google explores new infrastructure, Microsoft is changing how users interact with AI through its Copilot products.

On September 25, Microsoft announced a redesigned Copilot experience organised around three capabilities: Home, Code and Autopilot. The company described the changes as an effort to bring AI assistance, task execution and software creation into a more integrated environment.

Home combines Chat and Cowork in a central workspace. Chat is intended for everyday assistance, while Cowork supports delegating broader tasks that may require multiple steps.

Code is designed to let users build applications and automations using natural language. Microsoft says the feature uses technology related to GitHub Copilot and is intended to help people create solutions within their organisation’s environment.

Autopilot is designed as a persistent, proactive AI agent that can continue working on tasks rather than responding only to individual prompts. Microsoft has presented it as a way to move beyond simple question-and-answer interactions toward delegated work.

Microsoft also announced deeper integration of Word, Excel and PowerPoint into Copilot. Some capabilities are being introduced through the Microsoft Frontier program and private previews, so access depends on the feature and its rollout status.

The broader direction is to reduce the need to switch between separate applications and AI tools. Instead of asking Copilot for instructions and carrying out every step manually, users could increasingly delegate a task and review the work the system prepares.

Copilot Moves Closer to Windows Files and Actions

Microsoft’s October Windows and Surface announcements also showed how Copilot could interact more directly with a computer.

In a demonstration reported by The Verge, Microsoft’s Copilot leadership showed an AI workflow for preparing tax documents. The system located files across folders, renamed and compressed them, and drafted an email containing the resulting package for an accountant.

Microsoft described this direction as Hybrid Intelligence, combining local and cloud-based AI capabilities to complete tasks. The demonstration illustrated how an assistant could coordinate several actions across a computer rather than merely return a written answer.

Microsoft also previewed a redesigned Windows search experience that allows users to perform certain quick actions directly from the search interface. Examples shown included changing dark mode, adjusting microphone volume and sending a text.

These capabilities could make AI more useful for everyday computer work, particularly tasks that involve finding documents, organising files or moving information between applications.

However, a demonstration does not mean every user can access every capability immediately. Microsoft has said the Copilot features will roll out over time, and the exact experience will depend on availability, device support and software configuration.

Access to local files and the ability to perform actions also introduce important questions about permissions and user control. An AI system that can organise documents or prepare messages must operate within appropriate access boundaries, and users need to understand what actions it can take and when it requires confirmation.

What the Two Developments Mean for the AI Industry

Google and Microsoft are pursuing different parts of the AI technology stack.

Google’s experiment focuses on the physical infrastructure needed to run AI. If future research shows that orbital systems can operate reliably and economically, they could eventually offer another option for powering selected computing workloads. For now, radiation, thermal management, communications and cost remain substantial challenges.

Microsoft’s work focuses on how people use AI once the underlying computing is available. By integrating Copilot with applications, files and operating-system functions, the company is trying to make AI a more practical part of everyday work.

The initiatives also illustrate two different measures of progress. Google’s near-term success depends on whether its satellite can generate useful engineering data and validate assumptions about orbital hardware. Microsoft’s progress will depend on how reliably its AI tools complete real tasks, how widely the features become available and whether users trust them with sensitive work.

Neither development eliminates the need for conventional computing infrastructure or human oversight. Space-based AI remains experimental, and autonomous desktop assistance must still handle errors, permissions and sensitive information carefully.

What Happens Next?

Google’s next priority is to gather data from its orbital prototype and use the results to refine future designs. Further work on power systems, communications and satellite coordination would be necessary before an orbital network could support larger-scale AI computing. The project remains exploratory, and no commercial timetable has been established for replacing or materially shifting today’s data-center workloads into space.

Microsoft’s Copilot roadmap is focused on gradually bringing its new Home, Code and Autopilot experiences to more users, alongside Windows and Microsoft 365 improvements. The key test will be whether these features can complete useful multi-step tasks reliably while keeping users informed and in control. Together, the developments show how AI progress is taking place both in the infrastructure that powers models and in the software through which people use them.

Frequently Asked Questions

1. What is Google Project Suncatcher?

Project Suncatcher is Google’s research initiative exploring whether AI computing infrastructure could eventually operate in space. Its prototype satellite carries TPUs to test their performance under orbital conditions.

2. What are TPUs?

Tensor Processing Units are specialised processors developed by Google to accelerate machine-learning calculations. They are used in AI workloads and are the central computing hardware being tested in the Project Suncatcher prototype.

3. What are Microsoft’s new Copilot Home, Code and Autopilot features?

Home brings Chat and Cowork together, Code helps users create applications and automations using natural language, and Autopilot is designed to handle delegated tasks more proactively. Availability varies by feature and rollout stage.

4. Will space-based AI data centers replace those on Earth?

There is no evidence that this will happen soon. Google is testing the concept, and major technical and economic challenges remain. Space-based computing could eventually supplement terrestrial infrastructure for selected workloads, but its commercial viability has not been established.