Google CEO Sundar Pichai says the company aims to send its first equipment into space as early as 2027 to begin building artificial intelligence data centers off Earth. The idea is part of “Project Suncatcher,” a long-term research initiative announced by Google that focuses on developing machine learning capabilities in space.
Pichai admits that the concept sounds “extremely ambitious” and even “crazy,” but argues that the enormous energy requirements and computing demands of next-generation AI make the plan increasingly logical. The company hopes to place a TPU — Google’s custom chip designed to accelerate neural network processing — in orbit as a first step.
Read also: Google Romania hosts the largest wearable development center in EMEA
The rationale is straightforward: space offers continuous, unfiltered solar energy, available 24 hours a day. Unlike terrestrial data centers that require land, complex construction and huge amounts of electricity, an orbital facility could drastically reduce energy costs and ease pressure on national power grids.
Google is not alone in exploring this direction. Elon Musk, Jeff Bezos and Sam Altman have all recently discussed the long-term potential of space-based AI infrastructure. Musk claims that Starship could deliver hundreds of gigawatts per year in solar-powered AI satellites — an extraordinary figure compared to the current 59-GW global data center capacity. Salesforce CEO Marc Benioff described space as “the cheapest place for data centers” due to its unlimited solar power and lack of battery requirements.
Some tech leaders even imagine extreme future scenarios, such as constructing a Dyson Sphere-like structure around the solar system to capture massive amounts of energy needed for AI. However, shifting AI development into orbit comes with serious risks, including adding to Earth’s already growing space-debris problem.
Photo:Tom’s Hardware
