Earlier, some technology companies tried to send Nvidia AI computing accelerator cards to space, relying on solar energy and the lower temperature of space for testing. Theoretically, space has inexhaustible solar energy resources to power power, and the lower temperature of space does not require the deployment of air cooling or water cooling like ground data centers.

Now Google seems to have a similar plan. In its blog, Google envisions this new method that can bypass the resource constraints of energy-intensive artificial intelligence data centers on the earth. It will mount Google's TPU tensor computing unit on a satellite and launch it into the sky. The project is called the Sunshine Catcher Project.
If this project can be successfully launched, it will essentially build a data center in space. In this way, Google hopes to use solar energy to generate electricity around the clock, so that unlimited clean energy can be obtained, and the funds saved can continue to focus on the development of artificial intelligence technology.
But what's the problem?
Google says its tensor computing units are installed on satellites and launched into space. These satellites are equipped with solar panels for uninterrupted power generation. Google says solar panels in space are 8 times more efficient than those on the ground.
However, these chip-equipped satellites need to maintain good communications. Compared with data centers on the ground, space data centers need satellite links to support transmission rates of tens of terabits per second. Maneuvering satellite constellations into tight formations can help achieve this goal, but too close a distance may also pose a risk of collision.
Google must also ensure that the TPU units can withstand higher radiation levels in space. Currently, Google has conducted radiation endurance tests on Trillium TPUs. The tests show that these TPU units can withstand a total ionizing radiation measurement equivalent to a 5-year mission life without permanent damage.
Launch and investment costs:
The cost of launching a large number of satellites into space is still quite high at this stage, but Google's cost analysis shows that by the mid-2030s, the energy cost of launching and operating a data center in space may be roughly the same as that of a data center of the same size on Earth.
To this end, Google will cooperate with Planet to launch several prototype satellites for testing before 2027, by which time the actual operation of these hardware in space can be tested.