Google wants to build a data center that harnesses the power of the sun — in space (www.businessinsider.com)

Google announced Project Suncatcher, a research “moonshot” to test running AI compute directly in space by launching satellites packed with its Trillium-generation Tensor Processing Units. CEO Sundar Pichai says Google will send two prototype satellites to low-Earth orbit in early 2027; the company already ran TPU radiation resilience tests in a particle accelerator and outlined a vision of fleets of solar-powered nodes linked by high-speed optical inter-satellite links. The pitch: orbiting data centers could tap abundant solar energy, dramatically reduce terrestrial electricity and water usage, and—if launch costs fall below an estimated ~$200/kg by the mid-2030s—potentially become cheaper than building equivalent capacity on Earth. Technically ambitious, the plan faces hard engineering hurdles including thermal management, long-term reliability in radiation-rich environments, sustainment and maintenance of distributed space hardware, and secure, low-latency data exchange between orbit and ground. Google’s announcement also accelerates a nascent industry trend: SpaceX and startups like Starcloud are exploring similar ideas (Starcloud recently flew a GPU-equipped satellite), and Elon Musk publicly signaled interest. If feasible, space-based ML compute could reshape how large-scale models are provisioned and decouple AI growth from terrestrial resource constraints, but practical deployment depends on launch economics, hardware ruggedization, and orbital networking breakthroughs.