The United States Air Force has entered a new agreement with Redwood City-based WindBorne Systems to purchase atmospheric data gathered by the company’s autonomous weather balloons. The contract, valued at up to $49 million, operates on a pay-as-you-go basis, allowing the service to request data for specific regions and times. To date the Air Force has placed orders worth $1.7 million, and WindBorne’s executives anticipate additional purchases as operational needs evolve.
WindBorne maintains a constellation of roughly 700 balloons deployed worldwide, each equipped with atmospheric sensors that transmit temperature, humidity, pressure and wind measurements. The raw observations are sold to commercial clients and simultaneously fed into the firm’s proprietary deep-learning weather forecasting engine. Founded in 2015, the company positions itself among the first to integrate empirical balloon data directly into a neural network model, aiming to improve forecast precision and timeliness.
The Air Force contract marks WindBorne’s largest government deal to date, surpassing earlier collaborations with the National Weather Service, the Air Force and the Navy. Co-founder and chief technology officer Andrey Sushko explained to TechCrunch that the award, which did not involve competitive bidding, reflects a government determination that WindBorne alone could meet the required global data-collection capability. He added that the funding will support constellation growth and further research and development of balloon hardware and performance.
In August 2026 the company announced a $37 million Series B financing round intended to scale its operations, a move that preceded the release of WeatherMesh 6, its latest forecasting model. WindBorne claims WeatherMesh 6 delivers higher accuracy than the traditional European Centre for Medium-Range Weather Forecasts (ECMWF) system, which relies on physics-based simulations. The improvement is attributed to transformer-style deep-learning architectures that can ingest decades of historical weather data and generate predictions more rapidly than supercomputer-driven methods.
Earlier this summer WindBorne became the first organization to dispatch an unpowered flight system,lacking any engine,into the eye of a typhoon, demonstrating the practical use of its AI-driven navigation. The balloons adjust altitude, bleed air or release sand ballast to ride wind currents toward target locations. Sushko told TechCrunch that the achievement required both accurate wind and weather models and a sufficiently dense constellation to ensure balloons are positioned where needed, marking one of the first operational uses of deep-learning for flight path decisions.
WindBorne’s contract arrives amid a broader shift toward AI-enhanced meteorology, where transformer models and massive public datasets have enabled end-to-end learning that rivals traditional physics-based forecasts. Government and industry labs are producing increasingly accurate models; for example, Google DeepMind’s recent weather system adds cyclone-forecasting capabilities that are becoming central to U.S. hurricane response planning. While deep-learning and conventional approaches each retain distinct advantages, the convergence suggests a growing reliance on data-rich, machine-learning solutions for operational weather intelligence.