WindBorne Systems, a San Francisco-based startup founded in 2019 by Stanford graduates, has launched WeatherMesh 6, a new artificial intelligence weather forecasting model that outperforms the European Centre for Medium-Range Weather Forecasting (ECMWF), currently considered the global leader. The release marks a significant milestone where the startup claims superior accuracy in predicting key variables such as surface temperature. According to Chief Product Officer Kai Marshland, the model's five-day forecast is as accurate as traditional models are for the day before. Traditional weather prediction relies on complex physics simulations requiring expensive supercomputers and typically run every six hours. In contrast, WindBorne's AI-driven approach generates forecasts every hour with a resolution of just three kilometers across Europe and the continental United States. While other AI labs have produced faster models, many previously lacked the resolution or long-term accuracy of physics-based systems. WindBorne attributes its edge to a unique strategy that combines proprietary data collection with advanced deep learning. The company operates a fleet of approximately 400 weather balloons from 15 global sites, capturing direct sensor readings. The core advantage lies in data assimilation, the process of integrating disparate sensor inputs into a coherent model. While most AI weather systems depend on data sets provided by established agencies like the ECMWF and the US National Oceanic and Atmospheric Administration (NOAA), WindBorne feeds its data directly into its transformer-based architecture. CEO John Dean noted that the company has spent a year tuning and re-architecting the model to ensure stability and accuracy without relying on external initial conditions. Dean expressed confidence that removing ECMWF data would still result in a high-performing forecast. The technology has already found a foothold in government operations. WindBorne sells its raw balloon data to NOAA for use in the American weather forecasting enterprise, as well as to the US Air Force and Navy. Additionally, it markets its forecasts to commodity traders and investors. The company recently secured $25 million in venture funding and reached a valuation of $85 million in 2024. Safety remains a priority for the company. Last year, one of its balloons collided with a United Airlines jetliner, causing minor damage but no injuries, largely because the equipment adhered to US size regulations. In response, WindBorne has installed transponders on its balloons to transmit location data via the global aviation surveillance system, ADS-B, to prevent future collisions. Looking ahead, WindBorne is prioritizing the development of its model and data infrastructure over building consumer software products. CEO Dean emphasized a strategic shift away from traditional SaaS offerings, anticipating that future information consumption will be handled by AI agents rather than human users interacting with dashboards. This approach positions the startup to adapt to an evolving information environment where direct, high-frequency data integration is becoming the standard.