Starcloud Inc., a startup that hopes to build a 5-gigawatt artificial intelligence data center in orbit, has closed a $170 million funding round.

The company raised the capital in two tranches. Benchmark led the first transaction and co-led the second with EQT. The firms were joined by more than a half-dozen others, including Monolith Power Systems Inc., a publicly traded maker of power management chips.

The raise, which values Starcloud at $1.1 billion, comes about five months after it launched its first satellite. Starcloud-1 is perched in low-Earth orbit with an onboard H100 graphics processing unit. The company has used the chip to train a custom AI model and run an algorithm from Google LLC’s Gemini neural network series.

Starcloud says that space-based AI infrastructure offers several advantages over the standard terrestrial variety.

Data center cooling systems rely on a medium such as air or water to conduct heat away from servers. There’s no such medium in the vacuum of space. As a result, the only way to remove heat from a GPU is to radiate it in the form of infrared light.

Radiative cooling presents challenges that data center operators don’t encounter on Earth, but also has major benefits. The radiators used for the task are passive devices, which means they have a simpler design and use less power than terrestrial cooling systems.

Starcloud plans to power its space-based data centers with solar panels. On Earth, the amount of electricity generated by a solar panel fluctuates significantly based on the time of day and weather conditions. There are no such variables in orbit, which enables orbital solar panels to generate five times more power.

Starcloud is currently working on a successor to the satellite that it launched last year. Starcloud-2, as the new system is called, will feature what the company describes as the largest “commercial deployable radiator” ever launched to space. Its onboard solar array will generate 100 times more power than the one on the Starcloud-1.

The company plans to launch the satellite later this year. Once in orbit, Starcloud-2 will run AI workloads for multiple customers including Crusoe Inc., an AI data center builder.

Starcloud’s long-term vision is significantly more ambitious. The company hopes to assemble a 5-gigawatt orbital data center from modules that it calls containers. Each container will host multiple liquid-cooled AI servers along with network and storage equipment. A port with up to thousands of fiber-optic cables will enable the host container to exchange data with the other modules.

“By moving AI compute to space, we unlock access to unlimited solar power and completely remove the energy bottleneck,” said Starcloud co-founder and Chief Executive Officer Philip Johnston.

Starcloud plans to power its data center with a 6.1-square-mile solar array. According to a paper on the company’s website, laser beams or “small docking modules launched from the ground” could be used to upload training datasets. Starcloud estimates that the latter technology would make it possible to upload petabytes of data per trip.

In the more immediate future, the company will use its newly raised capital to build a manufacturing facility and grow its headcount. Starcloud also plans to book capacity on space launch vehicles.

Image: Starcloud