The U.S. Air Force has awarded WindBorne Systems a contract to buy atmospheric data collected by the startup’s fleet of autonomous weather balloons. The agreement can be worth up to $49 million, but the Air Force pays only as it needs data and can place orders for specific regions and times.
WindBorne's network and operations
WindBorne Systems, founded in 2015 and based in Redwood City, California, currently operates roughly 700 balloons worldwide. Those balloons collect atmospheric sensor readings that the company sells to customers and also uses to train and run its own AI-based forecasting model, WeatherMesh. WindBorne is among the first organizations to feed empirical measurements directly into a deep-learning weather model.
Contract details and current orders
The contract’s ceiling is $49 million; to date the Air Force has placed $1.7 million in orders under the agreement. Because the contract allows on-demand purchases for specific areas and times, WindBorne executives expect additional orders going forward.
Co‑founder and CTO Andrey Sushko said the award is largely a vehicle to support continued growth of the constellation and to fund research and development to improve balloon hardware and performance. The contract was awarded without competitive bidding; according to WindBorne, the government determined that for this capability of collecting global data, WindBorne was the only company able to meet the operational needs.
Government partnerships and funding
WindBorne has worked closely with U.S. government agencies including the National Weather Service, the U.S. Air Force and the U.S. Navy. In August, CEO John Dean told TechCrunch that the company had closed a $37 million Series B round to scale its business further.
AI in weather forecasting
WindBorne recently released WeatherMesh 6, and the company says this version provides more accurate forecasts than the traditional, physics-based model run by the European Centre for Medium-Range Weather Forecasts (ECMWF). Since transformer architectures and the ECMWF’s release of a large weather dataset a few years ago, AI-driven weather forecasting has become an increasingly productive application of deep learning.
Traditional meteorology relied on expensive supercomputers running detailed equations to simulate the atmosphere. End-to-end learning approaches can ingest decades of observational data and produce predictions more quickly. Models developed by leading labs and government agencies are proving in many cases to be more accurate than traditional forecasts, though the field remains a mixed landscape where both approaches are useful.
Navigating storms without engines
In August, WindBorne became the first organization to send an unpowered flight system — a vehicle with no engine — into the eye of a typhoon to collect data. The company navigates its balloons by changing altitude, bleeding air, or dropping sand ballast, allowing winds to carry the vehicles where they need to go.
Sushko described the capability as novel and said it rests on two ingredients: having accurate wind and weather models and knowing how to navigate them, and having a sufficiently large constellation so balloons are in the right places. He noted this was among the first instances where a deep-learning model was used to make actual operational navigation decisions.
Significance
The contract and the associated support for constellation growth should increase the volume of real-time atmospheric observations available to government and commercial forecasting systems. Incorporating such data into AI-based models can enable faster and, in some cases, more accurate forecasts — a particular benefit for predicting extreme weather and improving emergency preparedness.
