The U.S. Air Force will acquire data collected by WindBorne Systems' network of autonomous weather balloons under a newly established contract. This agreement could be worth nearly $50 million for the startup, marking a major milestone in its commercial and government partnerships.
Global Constellation and Advanced Weather Sensing
WindBorne currently operates approximately 700 weather balloons flying across the globe, gathering atmospheric sensor readings that the company commercializes for customers while also feeding them directly into its proprietary AI weather forecasting model. Based in Redwood City, California, and founded in 2015, WindBorne stands out as one of the very first organizations to feed empirical atmospheric data straight into a deep-learning forecasting model.
Government Collaboration and Contract Structure
While the company has previously collaborated closely with U.S. government entities like the National Weather Service, the U.S. Navy, and the U.S. Air Force, this agreement represents its largest participation contract to date. The total contract value reaches up to $49 million, though the Air Force operates on a pay-as-needed basis. It can place targeted orders for specific regions at exact times. So far, the agency has ordered $1.7 million worth of data, and leadership anticipates further orders down the line.
WindBorne co-founder and CTO Andrey Sushko noted that this arrangement primarily serves as a vehicle to support the continued expansion of the balloon constellation, alongside research and development efforts aimed at improving hardware and overall performance. Awarded without competitive bidding, Sushko explained that the government determined WindBorne was the sole company capable of providing global data collection in a manner that met their operational requirements.
Funding Milestones and WeatherMesh 6
WindBorne CEO John Dean noted earlier in August that the company successfully raised a $37 million Series B funding round to accelerate business scaling. Earlier in the summer, the startup rolled out the latest iteration of its forecasting model, WeatherMesh 6. According to the company, this model delivers superior forecast accuracy compared to traditional physics-based simulation models run by the European Centre for Medium-Range Weather Forecasts (ECMWF).
The Rise of AI-Driven Weather Forecasting
Following the emergence of transformer model architectures and the release of massive historical weather datasets by the ECMWF, artificial intelligence weather forecasting has evolved into one of the most productive applications of deep learning. Historically, meteorologists depended on expensive supercomputers executing complex equations to approximate atmospheric behavior. Today, end-to-end learning models ingest decades of weather records to generate rapid predictions.
Frontier labs and government agencies are demonstrating greater accuracy than conventional forecasts, though it remains a complex landscape where both forecasting methodologies are necessary for comprehensive weather understanding. Google DeepMind's latest weather model introduces critical new functionalities, while its cyclone forecasting tools have grown increasingly central to U.S. government hurricane response operations.
Navigating Balloons Into the Eye of a Typhoon
In August, WindBorne achieved a notable breakthrough by becoming the first organization to direct an unpowered flight system without an engine to collect atmospheric data straight from the eye of a typhoon. The company steers its balloons by dynamically adjusting altitude, venting air, or dropping sand ballast to catch specific winds that carry the vehicles toward their destinations.
Sushko emphasized that this represents a novel milestone, driven by two key ingredients: possessing accurate wind and weather models alongside expert navigation capabilities, and operating a constellation large enough to ensure balloons are positioned precisely where needed. This marks one of the earliest instances where a deep-learning model directly informed operational field decisions.
