Tech Firms Forge New Frontier with First Orbital Data Centers Powered by Nvidia H100 GPUs

October 28, 2025


A pioneering collaboration between space technology company Starcloud and cloud computing firm Crusoe is set to launch the world's first data centers into orbit, marking a significant leap in high-performance computing. The initiative will see Nvidia's powerful H100 GPUs operate in space for the first time, with the goal of harnessing unfiltered solar radiation to achieve gigawatt-scale computing capacity.


The partnership aims to move computational infrastructure off-planet to capitalize on the abundant solar energy available beyond Earth's atmosphere. Starcloud, a Redmond-based firm backed by the Nvidia Inception program, is developing the scalable orbital platforms. Crusoe will deploy its Crusoe Cloud platform on one of Starcloud’s satellites, which is scheduled for launch in 2026.


The core concept behind the venture is the efficient conversion of direct sunlight into computational power. Free from the filtering effects of the atmosphere, solar panels in orbit can capture sunlight at full intensity. The companies assert this will provide "almost unlimited, low-cost renewable energy" for demanding AI workloads. Initial estimates project that such space-based infrastructure could reduce energy expenses by up to ten times compared to terrestrial data centers, even after factoring in launch costs.


A key milestone is planned for November 2025, when Starcloud will launch the first Nvidia H100 GPUs into space. By early 2027, Crusoe expects to offer limited GPU-based computing power from orbit, describing the project as “a new paradigm for AI factories.” The plan involves running AI models and large language models (LLMs) directly on the orbital GPUs via Crusoe's cloud platform.


Proponents highlight additional potential benefits, including faster access to solar energy and more predictable thermal conditions. Nvidia has suggested the vacuum of space could act as "an infinite heat sink," enabling unconventional cooling methods not feasible on Earth.


However, the ambitious project faces significant challenges. The long-term feasibility remains uncertain, with questions surrounding the risks of space launches, hardware maintenance in orbit, and radiation exposure. The absence of traditional convection in zero gravity also presents a unique engineering hurdle for thermal management. While the vision for orbital data centers is clear, their practical and economic viability will ultimately depend on overcoming these complex obstacles.


SOURCE techradar.com

Read Also
Submer Launches InferX, a New Sovereign AI Cloud Division
Nvidia Announces $1 Billion Strategic Investment in Nokia to Accelerate AI in Telecom Networks
Google Inks Deal to Revive Iowa Nuclear Plant, Powering AI Expansion with Carbon-Free Energy

Research