Skip to content
01 of 05Data Centers in Space

Data Centers in Space

Library

Series and technical notes.

You are in Data Centers in Space · Introduction.

Series

Data Centers in Space

5 items

Watch video, summary and related content

Published General ~40 min 4 chapters
PrerequisitesAI and Generative AI Foundations

Compute demand is growing faster than the capacity to build data centers on the ground. This series asks whether moving computing infrastructure into space is a viable solution or a speculative bet: which physical problems vacuum and orbit actually solve, which ones they create, and which projects are currently demonstrations, proposals or early tests.

ground-orbit tradeoff
Orbit changes energy and water constraints, but it does not eliminate heat or mass
Comparing ground and orbit means separating genuine advantages from physical limits: grid capacity, water, links, radiators and every kilogram launched.
GroundOrbitThe question is not “ground or space”, but which bottleneck changes and which new limit appearsGrid: 4–10 yearscapacity constrainedLocal water2M L/day at 100 MWPermits + landviable sites take timeSun 95–99%for suitable orbitsHeatradiators 5–10 kg/m²Mass + linkevery kg and bit matters
Orbit trades grid and water constraints for more continuous solar energyIt does not erase heat: heat becomes radiator areaLaunch mass and communications still close the balance

Contents

1. Why now

  • AI compute demand has increased roughly 350,000× since 2014, while electricity scenarios are already forcing an infrastructure conversation (Chapter 1).
  • Six terrestrial bottlenecks recur: power grid, water, land, permits, heat and latency, with very different constraints depending on the case (Chapter 1).
  • Launch cost has fallen from roughly $88,000/kg for the Space Shuttle to around $1,400–2,500/kg for current reusable launchers; a <$200/kg threshold remains a projection, not an observed price (Chapter 1).
  • The current inflection point is still a mix of aggressive industry theses and regulatory proposals rather than a settled mass deployment (Chapter 1).

2. Energy, heat and connectivity

  • Why "space is cold" does not mean free cooling: heat rejection still dominates and radiator area grows quickly (Chapter 2).
  • A real advantage is longer-duration solar exposure in suitable orbits; the projected $0.002/kWh scenario remains an industry estimate, not an observed operating cost (Chapter 2).
  • Link windows, latency and downlink constraints mean orbital computing may improve some use cases without replacing terrestrial fibre (Chapter 2).
  • Orbital degradation makes maintenance difficult and pushes the architecture toward autonomy, redundancy and error correction (Chapter 2).

3. What a "data center in space" actually is

  • Real hardware already in orbit spans very different maturity levels, from satellite edge processing to early compute and storage demonstrators (Chapter 3).
  • The use-case spectrum ranges from useful onboard processing today to general-purpose cloud computing that is still speculative (Chapter 3).
  • Resilient storage and high-capacity nodes are plausible niches, but they are far from an orbital cloud equivalent to terrestrial infrastructure (Chapter 3).
  • The 1967 Outer Space Treaty remains foundational while questions around orbital digital sovereignty remain open (Chapter 3).
  • Most megaprojects are still moonshots, regulatory requests and company roadmaps rather than validated mass infrastructure (Chapter 3).

4. The real footprint of a data center

  • Water: national aggregates provide context, but environmental and political conflicts are local and depend heavily on cooling architecture (Chapter 4).
  • Energy: aggregate TWh is only part of the problem; rack-level power density increasingly determines facility design (Chapter 4).
  • Minerals: cobalt, rare earths, tantalum and copper add geopolitical and human dependencies that public debate often hides (Chapter 4).
  • Lifecycle: circularity helps but does not eliminate new chip demand or the material footprint that an orbital system would also have to launch (Chapter 4).

Related series: AI, GDP, Well-being and Energy · AI and Generative AI Foundations

View all series

Keep learning
Next chapterWhy nowData Centers in Space
All series →