Running artificial intelligence (AI) models incurs significant expenses, with a considerable portion of these costs stemming not from computing power alone but from the necessity of keeping servers cool. On Earth, this cooling entails the use of air conditioning, water systems, and substantial real estate. However, in low earth orbit, the surrounding temperature drops to nearly absolute zero, allowing heat to dissipate efficiently.
In a recent episode of The Samurai Code, Srinath Ravichandran, co-founder of Chennai-based Agnikul Cosmos, advocates for the concept of orbital computing. This idea is anchored on reusing a component of rockets that is typically discarded—the upper stage of the Agnibaan rocket, which has already proven capable of reaching orbit. This upper stage is being repurposed as a platform for GPU stacks in space. This approach effectively reduces costs in two significant ways: it eliminates the need for a separate satellite chassis and allows for a lighter payload, ultimately leading to lower launch expenses.
Ravichandran indicates that commercial availability of this technology could be realized within three to four years, contingent upon the development of space-hardened GPUs and a marked decrease in launch costs.
