In a move that seems inspired by science fiction, AI infrastructure firm Crusoe and satellite data center pioneer Starcloud announced a partnership to build and operate AI-powered data centers in orbit, leveraging the unlimited solar energy of space to power high-performance NVIDIA GPUs.
The first step comes this November, when Starcloud will launch its Starcloud-1 satellite equipped with NVIDIA’s H100 GPUs, the first data-center-class GPUs to operate in space. The 60-kilogram satellite, roughly the size of a mini-fridge, will provide 100 times more GPU compute power than any previous space-based mission.
A second phase will follow in 2026, when Crusoe deploys its Crusoe Cloud platform aboard a Starcloud satellite. Limited space-based GPU capacity is expected to be available to enterprise customers by early 2027.
Infinite Solar Power
Starcloud, to be sure, has an ambitious goal: to build a network of orbital data centers that reduce energy consumption on Earth. Traditional data centers depend on land, water, and grid electricity—resources that are fast becoming strained by AI’s exponential demand.
In orbit, however, constant sunlight provides effectively infinite solar power, and the vacuum of space acts as a natural cooling system. By radiating heat directly into the deep cold of space, Starcloud avoids the massive water use required for terrestrial cooling towers.
“In space, you get nearly limitless, low-cost renewable energy,” said Philip Johnston, Starcloud’s CEO. “The only cost on the environment will be on the launch, then there will be 10x carbon-dioxide savings over the life of the data center compared with powering the data center terrestrially on Earth.”
Johnston estimates that by 2035, most new large-scale data centers could be built in orbit, dramatically cutting global computing’s environmental footprint.
For Denver-based Crusoe, the venture represents a natural extension of its business model. The company built its reputation by co-locating compute infrastructure with underutilized or stranded energy sources such as flared natural gas. Now it plans to take that same energy-first strategy beyond Earth’s atmosphere.
Crusoe’s cloud platform will run AI workloads directly aboard Starcloud’s orbital data centers. The companies say the approach will enable new classes of near-real-time applications, such as processing satellite imagery for wildfire detection, crop monitoring, and emergency response. Because the data is processed in orbit—close to where it’s collected—response times could shrink from hours to minutes.
The NVIDIA Connection
Both startups are closely tied to NVIDIA. Starcloud is a member of the NVIDIA Inception startup accelerator program, which provides technical resources to promising AI companies. The firm’s first launch will showcase the NVIDIA H100’s capability to operate in extreme conditions, while future satellites are expected to use NVIDIA’s upcoming Blackwell platform, offering up to tenfold better AI performance.
Starcloud, which also graduated from the Google for Startups Cloud AI Accelerator, expects to use H100 GPUs to run Gemma, an open, lightweight large language model from Google.
NVIDIA, for its part, describes Starcloud’s project as “a milestone in the evolution of accelerated computing.” The company touts space-based computing as a powerful complement to its push for energy-efficient AI factories on Earth, and of course it also expands the potential market for its semiconductors.
A New Phase?
Building data centers in orbit poses daunting challenges, particularly radiation shielding. To help withstand this difficulty, each Starcloud unit will be a self-contained module—essentially a floating GPU cluster—with large solar arrays and an autonomous cooling system. Over time, the companies plan to scale the design to gigawatt capacities, forming what they call a “cloud constellation.”
If successful, the Crusoe-Starcloud partnership could herald a new phase in data center evolution. Just as early cloud computing abstracted away the need for local servers, orbital computing could abstract away the need for Earth-bound infrastructure altogether.
