Google DeepMind has introduced AlphaEarth Foundations, an artificial-intelligence model that combines petabytes of Earth-observation data into what the company calls a “virtual satellite.” The system weaves optical imagery, radar, lidar and climate simulations into a unified digital representation of the planet, allowing users to generate on-demand maps for any location and date. AlphaEarth characterises land and coastal waters in sharp 10-by-10-metre squares, compressing information into embeddings that require 16 times less storage than previous approaches while cutting error rates by roughly 24%. The result, according to DeepMind’s technical paper, is a scalable platform for tracking environmental change at city-block scale without the need for constant satellite fly-overs. To spur external research, Google is releasing a Satellite Embedding dataset derived from AlphaEarth—covering 1.4 trillion footprints a year from 2017 onward—through Google Earth Engine. More than 50 organisations, among them the UN Food and Agriculture Organization, MapBiomas and the Global Ecosystems Atlas initiative, have already used early versions to monitor Amazon deforestation, classify unmapped ecosystems and refine agricultural forecasts. The launch positions AlphaEarth as a cornerstone of Google’s broader Earth AI programme, which also includes flood, wildfire and weather models. By lowering the cost and complexity of geospatial analysis, the company expects the technology to inform government policy, corporate sustainability efforts and academic research on issues ranging from food security to water management.