Startup Wants to Run AI Inference From Space
Summary
Orbital Inc., a Los Angeles-based startup backed by Andreessen Horowitz, has emerged from stealth to announce plans for building space data centers to power AI inference workloads. The company aims to address the escalating energy demands of terrestrial data centers by harnessing abundant solar energy in low Earth orbit. Orbital envisions a mesh constellation of up to 10,000 fridge-sized satellites, each equipped with a GPU server rack, solar panels, and radiative cooling panels, providing 100 kilowatts of power per satellite. A prototype launch aboard a SpaceX Falcon 9 is planned for 2027 to validate GPU operations and run commercial inference workloads in orbit. Orbital's strategy focuses on distributed, smaller satellites optimized for less compute-intensive inference tasks, targeting major AI model labs like OpenAI and Anthropic as customers.
Key takeaway
For investors evaluating infrastructure plays in the AI boom, Orbital's space data center concept presents a high-risk, high-reward proposition. While it addresses critical energy and capacity constraints with a novel approach, you should carefully weigh the significant engineering challenges like radiation degradation, thermal management, and costly maintenance against the ambitious timeline and potential for long-term scalability. Your due diligence should focus on the feasibility of their planned 2027 prototype launch and the robustness of their radiation hardening and cooling solutions.
Key insights
Space-based data centers powered by solar energy aim to address terrestrial energy constraints for AI inference.
Principles
- Distributed inference is more feasible for space data centers.
- Solar energy in orbit offers abundant, "free" power.
Method
Orbital plans a mesh constellation of small satellites in low Earth orbit, each with a GPU server rack, solar panels, and radiative cooling, to process AI inference requests routed from ground stations.
In practice
- Consider radiation hardening for orbital GPUs.
- Implement liquid cooling for thermal management in vacuum.
- Focus on inference workloads for distributed satellite compute.
Topics
- AI Inference
- Space Data Centers
- Low Earth Orbit Satellites
- Solar Power
- Radiative Cooling
Best for: Investor, AI Architect, Director of AI/ML, Entrepreneur
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by IEEE Spectrum.