Juan A. Fraire

Researcher in Space Networking and Informatics

Inria CONICET Saarland University
Earth at night ringed by thousands of satellites in dawn-dusk orbits

Back to the home page

01 · RESEARCH TOPIC

Orbital Computing

Computers in orbit can process data where it is collected and serve people far from any ground station. They also cannot be repaired, every kilogram has to be launched, and all their heat must be radiated into space. I measure where orbital computing pays off and what it costs.

Satellites have carried computers for decades. What is new is the scale: in 2026, companies filed plans for constellations of data centres in orbit, from tens of thousands of satellites to as many as a million, powered by sunlight around the clock. The image above shows 13,020 of them, a sample I modelled on one of those filings.

The idea has real strengths, and some costs that are easy to leave out of a pitch. With colleagues at Saarland University, Nanjing University and the University of Genoa, I put numbers on both: latency, energy, mass and carbon over the whole life of the hardware, from the factory to re-entry.

Size a computing node

One node: an H100-class GPU with its controller, storage and laser link, 1 kW in all, in a dawn-dusk orbit at 510 km. Choose how long it flies, how many copies of the computer it carries, and which rocket launches it.

The numbers behind this: carbon per GPU-hour (g CO₂e)

From Fig. 1 of Dark Clouds Rising in Low-Earth Orbit (LEO-NET 2026), computed with the open-source ESpaS-ODC model. Rows: years in orbit. Columns: copies of the computer (1 = no spare). For comparison, a global-average data centre emits about 477 g per GPU-hour and one on a clean grid about 107 g.

Starship
Years12345
14416167919661141
2221308396484571
3147206264323381
4111155198242286
589124159194229
674103132162191
76489114139164
85678100121143
9506989108127
Falcon 9
Years12345
152971089210731255
2265355446537628
3177237298358419
4133178223269314
5106143179215252
689119149180210
776102128154180
86789112135158
95980100120140

The radiator is sized by power, at 3.03 m² per kilowatt (calibrated on the International Space Station), so mission length and spares leave it unchanged. A GPU in orbit is expected to last about three years, so a mission of T years needs at least ⌈T/3⌉ copies. A fuller calculator built on ESpaS is planned.

Where it can work, and what it costs

Where it can work

  • Distance to the ground. A cloud in orbit answers faster than one on the ground when the user is more than about 5,000 km from the nearest gateway. Under 2,000 km, the ground wins. ASMS/SPSC 2025, §V-A
  • Shrinking data before sending it. Processing in orbit saves carbon when it cuts the data enough: with one inter-satellite hop, down to 50% of the raw size on Falcon 9, or 70% on Starship. Dirty Bits, §4.3
  • Against an average data centre. Within its first one or two years in orbit, a node emits less carbon per GPU-hour than the global-average data centre on Earth (about 477 g of CO₂e). Dark Clouds, LEO-NET 2026, Fig. 1

What it costs

  • Heat. Every watt a computer draws ends up as heat, and in vacuum it can only be radiated away. For a 1 kW node, the radiator weighs 30.5 kg of the 44.3 kg modelled; the computer itself, 3.2 kg. Dark Clouds, §4 Scaled to the 250 kW that SpaceX quotes for each of its Star Mind spacecraft, the same ISS-based density gives about 760 m² and 7.6 tonnes of radiator, the heat load of some 18 ISS radiator panels. Dark Clouds, §3
  • Launch and re-entry. Each kilogram costs about 52 kg of CO₂e on Falcon 9 and 34 on Starship. Re-entry adds about half again on top of the launch. Dirty Bits, §3.1
  • No repair. Spares have to fly from day one. One cold spare of the computer adds 40% to the carbon per GPU-hour on Starship, and 34% on Falcon 9, for a three-year mission. Dark Clouds, abstract
  • Against a clean data centre. Matching a data centre on a clean grid (about 107 g per GPU-hour) takes at least six years in orbit with one spare on Starship, or nine years with two spares on Falcon 9. Dark Clouds, §4

Carbon figures count manufacturing, launch and re-entry of the modelled components; once in orbit the node runs on sunlight. The satellite bus (structure, attitude control, avionics) is left out, as in the papers; including it would add mass.

Tools and projects

Selected papers

  1. Dark Clouds Rising in Low-Earth Orbit: On Environmental Limits to Massive Orbital AI Robin Ohs, Gregory F. Stock, Andreas Schmidt, Juan A. Fraire, Jörg Ott, and Holger Hermanns ACM Workshop on LEO Networking and Communication (LEO-NET), co-located with ACM SIGCOMM, 2026
    Abstract

    In 2026, we saw rising numbers of proposals for orbital data centers to facilitate AI, e.g., by SpaceX (up to one million satellites), Blue Origin, and Google—yet none include rigorous lifecycle sustainability analysis. We present ESpaS-ODC, a lifecycle carbon model that accounts for the three subsystems physically unavoidable at GPU-class power densities but absent from prior work: a thermal radiator sized from ISS data, solar-array degradation and eclipse margin, and cold-standby spares for no-repair access. Applying the model to an edge data center in a 510 km orbit reveals that the service-overhead-scaled radiator weighs about eleven times the GPU it serves and that modeled power and thermal infrastructure dominate launched component mass—not compute. Using mission duration T as the denominator to amortize launch, re-entry, and manufacturing carbon, model-boundary parity with a global-average terrestrial DC occurs within the first two mission years in our parameter sweep. With respect to a renewables-powered green-energy DC (Finland grid intensity: 68 gCO2e/kWh, PUE: 1.2), parity requires multi-year missions with cold-standby spares compensating the shorter component lifetimes. Idealized dawn-dusk SSO (no-eclipse regime, β > βcrit) eliminates the eclipse battery, cutting modeled component mass significantly (e.g., 25% on the 1 kW-ODC for a 3 yr mission), but leaves the radiator unchanged: high-beta orbits solve the battery problem, not the thermal problem. Finally, carrying one full cold spare (r = 2) increases amortized carbon per GPU hour by 40% on Starship and 34% on Falcon-9 at T = 3 yr, while its dependability benefit remains to be quantified.

  2. Optimal Design of Space-Based Data Center Constellations for In-Orbit Computing Services Wenxiao Ge, Juan A. Fraire, and Kanglian Zhao IEEE Communications Letters, 2026
    Abstract

    With the increasing demand for in-orbit computing services, the architectural design of Space-Based Data Center (SBDC) constellations has emerged as a critical research challenge. However, existing design methodologies remain tailored to traditional constellations and lack specialized models to evaluate the unique cost-performance trade-offs. This letter proposes a comprehensive multi-objective optimization model for SBDC constellation design, aiming to minimize cost and delay while maximizing computational performance. To efficiently solve this complex problem, a novel SBDC Constellation Design Optimization (SCDO) framework, built upon NSGA-II, is developed to obtain the optimal constellation architectures. Extensive simulations demonstrate the effectiveness of the proposed framework, reveal two primary design paradigms, and provide an in-depth analysis of the impact of task characteristics.

  3. Space Cloud Networks: Technologies, Architectures, and Key Enablers Wenxiao Ge, Juan A. Fraire, Xinxin Shen, and Kanglian Zhao IEEE Communications Magazine, 2026
    Abstract

    The rapid growth of artificial intelligence models and global computing demands has placed an unsustainable burden on ground-based data centers, exacerbating energy consumption and environmental concerns. Simultaneously, emerging space-based infrastructures face stringent real-time processing requirements, while underserved regions experience persistent connectivity gaps. To address these challenges, we propose a space cloud network powered by space-based data centers interconnected through high-speed laser links. We present a three-layer architecture comprising cloud, access, and user layers and analyze the strengths and limitations of different orbits and configurations. We advocate for a multi-layer heterogeneous fusion network that leverages dawn-dusk and inclined orbits to balance performance, latency, and coverage. Additionally, we identify key enabling technologies, including space cloud operating systems, file systems, network management, and network security. This work presents a comprehensive framework for establishing the space cloud network, aiming to enhance global connectivity and computational efficiency.

  4. Dirty Bits in Low-Earth Orbit: The Carbon Footprint of Launching Computers Robin Ohs, Gregory F Stock, Andreas Schmidt, Juan A. Fraire, and Holger Hermanns ACM SIGENERGY Energy Informatics Review, 2025
    Abstract

    Low-Earth Orbit (LEO) satellites are increasingly proposed for communication and in-orbit computing, achieving low-latency global services. However, their sustainability remains largely unexamined. This paper investigates the carbon footprint of computing in space, focusing on lifecycle emissions from launch over orbital operation to re-entry. We present ESpaS, a lightweight tool for estimating carbon intensities across CPU usage, memory, and networking in orbital vs. terrestrial settings. Three worked examples compare (i) launch technologies (state-of-the-art rocket vs. potential next generation) and (ii) operational emissions of data center workloads in orbit and on the ground. Results show that, even under optimistic assumptions, in-orbit systems incur significantly higher carbon costs—up to an order of magnitude more than terrestrial equivalents—primarily due to embodied emissions from launch and re-entry. Our findings advocate for carbon-aware design principles and regulatory oversight in developing sustainable digital infrastructure in orbit.

  5. On the Latency Trade-off Between Space and Terrestrial Clouds in Non-Terrestrial Networks Camilo Rojas, Juan A. Fraire, Fabio Patrone, and Mario Marchese Advanced Satellite Multimedia Systems Conference and Signal Processing for Space Communications Workshop (ASMS/SPSC), 2025
    Abstract

    Non-Terrestrial Networks (NTN) are poised to revolutionize 5G and 6G networks by integrating terrestrial and space-based cloud systems, enabling dynamic task allocation for optimal performance. Despite their promise, understanding the trade-offs between terrestrial and non-terrestrial edge computing architectures remains an area for improvement. This paper presents a comprehensive latency-focused trade-off analysis using a novel real-time emulation platform that accurately models terrestrial and space cloud environments. By evaluating network latency across geodesic distances from a fixed ground gateway, we delineate scenarios where terrestrial clouds excel and identify conditions under which Space Cloud architectures surpass their terrestrial counterparts. Additionally, we analyze how server placement strategies in satellite constellations impact performance, revealing the critical interplay between server distribution and latency outcomes. These findings offer actionable insights for designing and operating hybrid cloud systems, emphasizing the need for tailored architectures to maximize the potential of NTN-based edge computing.

All papers on orbital computing

People and partners

With Saarland University Nanjing University · University of Genoa · TU Munich Funding European Union MISSION (Marie Skłodowska-Curie RISE) DFG (TRR 248 CPEC)

Writing about orbital data centres?

I can explain the trade-offs, check numbers, or point you to the right people. The press kit has bios, a photo and these results with their sources.

juan.fraire@inria.fr