6 months ago
60
ARTICLE AD BOX
Google has announced a probe inaugural to physique AI information centres successful space, marking what the tech elephantine calls a caller moonshot that could alteration however instrumentality learning compute is scaled.
The project, called Suncatcher, envisions compact constellations of solar-powered satellites carrying Google's Tensor Processing Units (TPUs) and connected done free-space optical links. The announcement comes weeks aft Amazon laminitis Jeff Bezos predicted that gigawatt-scale information centres volition beryllium built successful abstraction wrong the adjacent 10 to 20 years.The institution announced the inaugural alongside a preprint probe insubstantial titled "Towards a aboriginal space-based, highly scalable AI infrastructure strategy design."
Google CEO Sundar Pichai says prototype motorboat successful 2027
Alphabet and Google CEO Sundar Pichai hailed the project, giving a timeline erstwhile the institution expected to motorboat 2 prototype satellites. Read his afloat post: Our TPUs are headed to space!.Inspired by our past of moonshots, from quantum computing to autonomous driving, Project Suncatcher is exploring however we could 1 time physique scalable ML compute systems successful space, harnessing much of the sun’s powerfulness (which emits much powerfulness than 100 trillion times humanity’s full energy production),” helium saLike immoderate moonshot, it’s going to necessitate america to lick a batch of analyzable engineering challenges. Early probe shows our Trillium-generation TPUs (our tensor processing units, purpose-built for AI) survived without harm erstwhile tested successful a particle accelerator to simulate low-earth orbit levels of radiation. However, important challenges inactive stay similar thermal absorption and on-orbit strategy reliability. More investigating and breakthroughs volition beryllium needed arsenic we number down to motorboat 2 prototype satellites with @planet by aboriginal 2027, our adjacent milestone of many. Excited for america to beryllium a portion of each the innovation happening successful (this) space!
Why abstraction for AI?
Google's thought centres connected vigor abundance and efficiency. The institution said that the Sun emits much powerfulness than 100 trillion times humanity's full energy production, making it the eventual vigor root successful our star system. In the close orbit, a star sheet tin beryllium up to 8 times much productive than connected Earth and nutrient powerfulness astir continuously, reducing the request for batteries.
“Artificial quality is simply a foundational exertion that could reshape our world, driving caller technological discoveries and helping america tackle humanity's top challenges,” Google stated successful its announcement, adding, “Now, we're asking wherever we tin spell to unlock its fullest potential.”The institution argues that abstraction whitethorn yet beryllium the champion spot to standard AI compute, portion besides minimising interaction connected terrestrial resources – an progressively important information arsenic AI's vigor demands grow.
Technical challenges ahead
The preprint insubstantial describes Google's advancement toward tackling foundational challenges, including high-bandwidth connection betwixt satellites, orbital dynamics, and radiation effects connected computing hardware.By focusing connected a modular plan of smaller, interconnected satellites alternatively than monolithic azygous structures, Google is laying the groundwork for what it describes arsenic a highly scalable, aboriginal space-based AI infrastructure.
