The platform turns petabytes of daily satellite imagery into real‑time, queryable intelligence, giving decision‑makers unprecedented speed and precision for monitoring global change.
Creon Levit, senior engineer at Planet Labs, presented the company’s AI‑driven Earth observation platform at Vision Weekend USA 2025, outlining the transition from its original daily‑coverage mission to a new “queryable planet” service that lets users ask natural‑language questions about the planet.
Planet Labs now operates more than 200 imaging satellites that capture roughly 90 % of the world’s land each day, downlinking about 20 TB of eight‑band, 4‑meter resolution imagery. The data are sliced into 256‑pixel tiles, producing roughly 260 million tiles per day, each embedded into a vector space using multiple multimodal large language models.
The embedded tiles are stored in a vector database and combined with text‑to‑image embeddings, enabling one‑second global searches for change between two dates. Live demos showed the system locating construction sites, highway interchanges, and cooling towers after only a few relevance‑feedback clicks, and even linking to Planet’s chat interface for additional context.
By turning raw satellite data into an instantly searchable knowledge base, Planet Labs offers governments, utilities, and enterprises a rapid way to monitor infrastructure, assess environmental impacts, and respond to emerging events, potentially reshaping the economics of Earth‑observation analytics.
Comments
Want to join the conversation?
Loading comments...