What happened

Google, the company behind Search and AI products, has begun testing an AI computing chip in orbit. The prototype satellite launched from California on October 1 aboard a SpaceX rocket, according to TechCrunch. The chip is a Tensor Processing Unit, or TPU. Google designs TPUs for AI calculations.

The test is part of Project Suncatcher. Google hopes to build networks of solar-powered satellites that share AI work. The first satellite is deliberately small. After activation, its TPU is expected to run in 15-minute bursts. This helps engineers study power use, heat, and model performance.

Why put AI in space?

Earth-based data centers need large amounts of electricity. They also need land and equipment to remove heat. Google says satellites could use sunlight for long periods. That could reduce pressure on some Earth-based resources. Its official explanation describes radiation, cooling, and communication as major engineering problems.

Space does not make cooling simple. Orbiting machines have no ordinary airflow. Engineers must move heat through radiators and other systems. Satellites also need power equipment, communication links, and protection from radiation. Every part must survive the launch and keep working far above Earth.

Google’s larger plan uses many satellites working together. Fast laser links could let them share data. That connection matters because large AI workloads divide work across many chips. One small test satellite cannot provide the scale of a normal data center.

What 1,800 launches means

The 1,800 figure comes from a cost model, not a launch order. Google’s research models launch prices falling close to $200 per kilogram by 2035. It assumes SpaceX’s rocket operations keep improving by about 20% each year.

Under that model, Starship would need to carry roughly 370,000 metric tons into orbit over ten years. If each mission carried 200 metric tons, the total would be about 1,800 launches. That means roughly 180 launches each year. The article’s web address says 1,600 launches, while its article text gives 1,800. The exact number is therefore only an estimate.

What is confirmed

Google has launched a prototype satellite carrying its advanced chip. The immediate goal is to learn whether the TPU can work in space. Engineers must measure power, heat, radiation effects, communication, and errors during real operation.

The report says Google’s long-term vision includes a constellation of 81 satellites. It also notes that Starship has never flown more than five times in one year. Reaching 180 launches per year would require a huge increase in launch frequency.

What remains unclear

We do not yet know how long the chip will operate, or how fast it will run useful models. We also do not know whether space computing will be cheaper, faster, or more reliable than computing on Earth. Google’s research suggests radiation may be manageable for ordinary inference work. Very large training jobs could be more sensitive to errors.

It is also unclear how closely the cost model matches real rocket performance. The 1,800 launches describe a future condition for lower costs. They do not prove that a full orbital data center is ready.

What to watch next

Google has discussed testing two more purpose-built satellites in 2027. Those satellites may test laser links between spacecraft. The next important evidence will be data about cooling, power, radiation, and communication. SpaceX’s launch rate and payload capacity will matter too.

For now, Google has tested one small part of a much larger idea. A space data center is not operating yet. The important change is that the idea has moved from a research design to a real orbital test.