Signal Jul 11, 2026 at 10:447Add to bookmarks

Several startups announce AI in low Earth orbit. IEEE Spectrum did the math - it's marketing, not infrastructure.
For eighteen months, several startups have presented their ambition to run AI in low Earth orbit: "free cooling by the vacuum, unlimited solar energy, global latency". IEEE Spectrum took stock on July 1, 2026. Verdict: no demonstration in orbit of a substantial computing load. The announced prototypes are a few watts - a fraction of a smartphone.
The vacuum is NOT a cooler. The absence of atmosphere means that heat can only be dissipated by radiation, drastically less effective than convection or liquid. Space solar is intermittent (Earth occultation with each orbit) and limited by the surface area of the panels. The orbit-to-ground link costs 10 to 100 times more watts per bit than a fiber backhaul. Physics says no; the pitch deck says yes. The story is used to raise funds on a "climate" ticket, not to deliver FLOPS.
The real fundraisings on this narrative. If a Tier-1 fund seriously finances beyond the demonstration check, the analysis will need to be updated. Meanwhile, to be filed under "vaporware 2026", alongside the corporate metaverse and the LLM with 100,000 billion parameters promised for next year.
Article produced by artificial intelligence, reviewed under human editorial control.
Sign in to join the discussion.
Qui contrôle ces data centers spatiaux ? Comment garantir la sécurité des données ?
Oui, il faut des règles claires pour éviter les abus. Comme pour les satellites, mais en plus strict.
Et les problèmes de latence ? Même en orbite basse, les délais de transmission ne justifient pas tout ce battage.
Et la redondance ? Une panne en orbite, ça peut être dramatique.
Est-ce que le coût du transfert de données vers et depuis l'espace compensera les gains de performance ?
La latence de transmission des données entre la Terre et l'espace, ça ne vous inquiète pas ?
Est-ce que ces startups innovent vraiment ou elles courent après la hype ?
Est-ce que le coût énergétique du lancement et de la maintenance de ces data centers dans l'espace compense les avantages du traitement en basse latence ?