How a balloon startup is beating Europe’s best weather forecasters with AI

Windborne Systems, a startup founded by Stanford alumni in 2019, has released the latest version of its AI weather model, WeatherMesh 6, and claims it now outperforms the gold standard in global forecasting—the European Centre for Medium-Range Weather Forecasting (ECMWF). The company says its model delivers more accurate predictions on key variables, including surface temperature, and does so at hourly intervals instead of the standard six hours. Resolution has improved to 3 kilometers over Europe and the continental U.S., where data quality is highest. What gives Windborne an edge is its unconventional approach: it builds both the hardware and the AI. The company operates a fleet of roughly 400 weather balloons, launched from 15 sites worldwide, that feed sensor data directly into its deep learning model. While most AI weather models still rely on government-produced datasets for initialization, Windborne has spent a year retuning its transformer-based architecture to ingest its own balloon readings. That direct data pipeline, says head of AI Joan Creus-Costa, is the main driver behind the accuracy gains. CEO John Dean puts it bluntly: “I don’t understand the business model of being an AI weather company without a data set advantage.” He notes that removing ECMWF’s initial conditions from WeatherMesh 6 would still yield strong results—a sign the model is becoming self-sufficient. The company sells its balloon data to NOAA, the U.S. Air Force, and the Navy, and offers forecasts to commodity traders and investors. But Dean is cautious about building a full SaaS product, given how quickly consumer information channels are shifting toward AI agents. Windborne has raised $25 million in venture funding, with a reported valuation of $85 million as of 2024. A minor incident last year—a United Airlines jet struck one of its balloons—prompted Windborne to add ADS-B transponders to its fleet for better visibility in aviation airspace. No injuries were reported. As weather AI models improve rapidly, Windborne’s hybrid strategy of owning both the data pipeline and the model could give it a lasting edge over competitors and government agencies alike.