<\/p>\n\n
For the past year, orbital data centers have been making headlines. Most announcements remain projections. But on October 1, Google launched the first experimental satellite for its Suncatcher project, and its researchers published their design in the journal Joule [1, 2].<\/p>\n\n
On Earth, building AI data centers faces obstacles related to grid connections, water supply, and public acceptance. Many projects are delayed or abandoned. In orbit, on a dawn-dusk sun-synchronous orbit, solar panels receive nearly continuous sunlight: Google estimates up to eight times more energy generated per panel per year than on Earth\u00b9.<\/p>\n\n
The race is underway. Starcloud placed an H100 GPU in orbit in November 2025\u2074. SpaceX is seeking authorization for up to one million satellites\u2075. Blue Origin, Axiom, and Aetherflux are also advancing their ambitions.La course est lanc\u00e9e. Starcloud a plac\u00e9 un GPU H100 en orbite en novembre 20254<\/sup>. SpaceX demande une licence pour jusqu’\u00e0 un million de satellites5<\/sup>. Blue Origin, Axiom et Aetherflux ont aussi des ambitions qui se concr\u00e9tisent.<\/p>\n\n Rather than a giant satellite, Suncatcher proposes a cluster of small satellites flying in formation. Its reference scenario involves 81 satellites at an altitude of approximately 650 km, spaced 100\u2013200 m apart and connected by lasers\u00b9. No performance results have been published yet. A two-satellite test to validate the optical links is planned for early 2027\u2076. Google\u2019s choice reflects an engineering rationale: distributing computing power across multiple satellites instead of concentrating it in a single structure. This modular approach allows heat to be managed at the individual-unit level, modules to fit within existing launch-vehicle fairings, and a failed satellite to be replaced without losing the entire cluster.<\/p>\n\n But that flexibility comes at a cost: the satellites must maintain precise formation and reliable optical communications. That is Suncatcher\u2019s challenge: demonstrating that a distributed architecture can operate efficiently in orbit.<\/p>\n\n 1. Heat. In a vacuum, heat can only be dissipated through radiation, requiring large radiator surfaces and therefore additional mass. Google\u2019s paper acknowledges this as a major challenge but does not quantify radiator area or mass\u00b9. An IEEE Spectrum analysis concludes that operating a GPU in orbit would cost at least ten times as much as on the ground, even under optimistic assumptions\u2077. Estimates of thermal-system mass vary by a factor of ten across sources\u2078.<\/p>\n\n 2. Launch costs. Google\u2019s economic case assumes launch costs below approximately $200\/kg by 2035, compared with around $3,600\/kg today on Falcon 9. Under its own model, this would require sustaining roughly 180 Starship flights per year\u00b9. And the comparison concerns energy costs, not the total cost of a data center.<\/p>\n\n 3. Reliability. In orbit, radiation poses a significant challenge for electronic components. While total radiation dose can be managed through component selection and shielding, energetic particles can cause transient errors, including bit flips. Google plans to mitigate their effects through hardware redundancy\u00b9 and software-based error correction. The challenge remains ensuring long-term system reliability without the possibility of direct repairs.<\/p>\n\n 4. Orbital congestion. Nearly all projects target the same dawn-dusk orbit. Applications explicitly filed for this orbit total approximately 111,000 satellites\u2079, in a band estimated to represent around 1% of usable low-Earth-orbit volume\u00b9\u2070. There are currently about 16,000 active satellites\u2079. Google estimates that terawatt-scale computing capacity could fit in this orbital band\u00b9, a claim challenged by 33FG\u2019s analysis\u00b9\u2070. SpaceX argues that its autonomous collision-avoidance maneuvers make this density manageable\u00b9\u00b9, but coordination remains an open question. With 81 satellites, Suncatcher is the most modest of the current projects. <\/p>\n\n – Short term: Orbital computing is most compelling for niche applications, such as preprocessing Earth-observation imagery before transmitting it to the ground.<\/p>\n\n – Long term: A general-purpose, gigawatt-scale data center remains an aspiration for around 2035 or later, contingent on launch costs that have yet to be demonstrated.<\/p>\n\n – What to watch: Prototype thermal data and error rates, the 2027 test, actual Starship launch pricing, and orbital regulation.<\/p>\n\n The authors themselves conclude that AI in space is promising but faces serious engineering challenges\u00b9. Google\u2019s main contribution is to put numbers, assumptions, and limitations on the table. That is what separates an R&D project from an investor presentation.<\/p>\n\n <\/p>\n\n <\/p>\n\n <\/p>\n","protected":false},"excerpt":{"rendered":" For the past year, orbital data centers have been making headlines. Most announcements remain projections. But on October 1, Google launched the first experimental satellite for its Suncatcher project, and its researchers published their design in the journal Joule [1, 2].<\/p>\n","protected":false},"author":9,"featured_media":5141,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1,123],"tags":[],"class_list":["post-5142","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-innovation-posts","category-news-2"],"acf":[],"_links":{"self":[{"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/posts\/5142","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/users\/9"}],"replies":[{"embeddable":true,"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/comments?post=5142"}],"version-history":[{"count":1,"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/posts\/5142\/revisions"}],"predecessor-version":[{"id":5143,"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/posts\/5142\/revisions\/5143"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/media\/5141"}],"wp:attachment":[{"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/media?parent=5142"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/categories?post=5142"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/innovitech.com\/en\/wp-json\/wp\/v2\/tags?post=5142"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}Why 81 Small Satellites Instead of One Large One?<\/strong><\/strong><\/h3>\n\n
Four Major Obstacles<\/strong><\/h3>\n\n
Key Takeaways<\/strong><\/h2>\n\n
References<\/strong><\/h2>\n\n
\n