The demand for computing power for AI is growing faster than the Earth can sustain. But a startup founded in Los Angeles wants to solve this problem by sending data centers into space. The idea received support from venture capital manager Andreessen Horowitz (a16z), which announced this Tuesday (14) an investment – of an undisclosed amount – in Orbital-1, Orbital’s first test mission.
The company wants to build and operate data centers in low Earth orbit. Each satellite will house a cluster of servers with Nvidia GPUs, powered by solar panels and cooled passively – simply radiating heat into the vacuum. In heliosynchronous orbit, the sun never sets, there is no dependence on the electrical grid, and cooling – one of the biggest costs of any terrestrial data center – happens for free.
“AI progress is being limited by the power grid,” says Euwyn Poon, CEO and founder of Orbital. “The economics of data centers are dominated by electricity and cooling, and both are getting tougher. In orbit, solar power is continuous and cooling is fundamentally different. Orbital is building a computing infrastructure that removes the power ceiling and scales with the potential of AI.”
Continues after advertising
This is not the first venture for Euwyn Poon, who previously founded Spin, an electric scooter startup acquired by Ford in 2018 for US$100 million.
Orbital’s first mission is scheduled for April 2027, aboard a SpaceX Falcon 9. The goal is to validate the continuous operation of GPUs in orbit, test the hardening of components against space radiation and, after validation, run real AI inference payloads in space. The company is also in the process of registering with the Federal Communications Commission (FCC) to obtain a license to operate a constellation of satellites.
In addition to the satellite, Orbital is opening Factory-1, its first R&D facility in Los Angeles, where satellites will be designed and manufactured in the United States.
a16z’s investment in Orbital was made through Speedrun, an intensive 12-week program that guides founders through each critical stage of their growth. “Speedrun supports founders to explore ambitious ideas – the harder the problem, the better,” said Andrew Chen, general partner at a16z Speedrun, in a statement. “Orbital is tackling AI’s biggest constraint with a bold and radical idea.”
The idea of launching data centers in space is not exactly new, and there are already some initiatives in this direction being developed. Google, SpaceX, Amazon and Nvidia have already publicly signaled interest in the topic. The startup Starcloud launched Starcloud-1 in 2025, equipped with an Nvidia H100 GPU, and reported having trained a language model in space. Lonestar last year announced plans to install a data center on the lunar surface.
Orbital’s computing infrastructure was designed based on specific technical understanding. Training large AI models requires thousands of tightly interconnected GPUs, communicating with near-zero latency. This architecture does not apply to satellites. The inference is different. Each request is handled independently and capacity can be distributed across multiple nodes. Orbital focuses on inference, where orbital computing can scale as a constellation and serve workloads in parallel.
Continues after advertising
Content produced by .
