Usman Salis
⬤ Google is quietly developing what sounds like science fiction yet could arrive sooner than expected - a fully operational AI data center that floats in space. The company intends to lift a supercomputing cluster into orbit freeing it from the limits of Earth's power grid and from environmental rules. The idea is no longer a daydream - detailed engineering plans exist and the Chinese company Comospace is reportedly pursuing a nearly identical project.
⬤ The design splits into three linked parts that would operate together in orbit. The first part is a 100-megawatt power module. It relies on huge solar panels and batteries - the platform draws steady electricity without any connection to Earth. The second part is a communication module that sends ten terabits per second through laser beams to other satellites and to ground stations. The third part is the computing core - tens of thousands of AI accelerator cards packed into a cluster that can deliver ten exa operations per second. Those numbers reveal the scale of hardware that firms like NVDA besides GOOG expect to deploy as AI workloads expand.
The system consists of three primary modules engineered to function together in space.
⬤ The parallel effort by China's Comospace shows that orbital computing is no longer a lone venture. The sketches that circulate depict high output solar arrays, laser links and dense compute trays sharing one orbital platform. No launch date has been announced - yet the depth of the blueprints signals a firm long term wager on shifting AI processing away from Earth.
⬤ The concept forces the industry to confront where AI infrastructure is heading. If data centers in orbit prove practical, demand patterns for NVDA chips could shift or GOOG would need to revise its capacity plans. Development is still at an early stage, but the simultaneous interest from multiple parties indicates that orbital computing may become the next arena for managing the rapid rise in AI workloads.
Usman Salis
