WindBorne Soars: US Air Force Awards Hefty Contract for AI Balloon Tech

The U.S. Air Force has secured a significant contract with WindBorne Systems, an innovative startup, to acquire data collected by its fleet of autonomous weather balloons. This agreement could be valued at nearly $50 million, marking the single-largest contract the company has participated in to date. WindBorne, established in 2015 and headquartered in Redwood City, California, operates approximately 700 weather balloons globally, gathering crucial atmospheric sensor readings.

The data collected by WindBorne's balloons is sold to various customers and is also integrated into the company's proprietary AI weather forecasting model. As one of the pioneering organizations to directly input empirical data into a deep-learning weather model, WindBorne has forged close ties with the U.S. government, including collaborations with the National Weather Service, the U.S. Air Force, and the U.S. Navy.

Under the terms of the Air Force contract, which is worth up to $49 million, payments are made on an as-needed basis, allowing the Air Force to place orders for specific regions and times. Currently, orders totaling $1.7 million have been placed, with executives anticipating further substantial orders. Andrey Sushko, WindBorne co-founder and CTO, explained that this contract primarily serves as a mechanism to support the continued expansion of their balloon constellation and to fund research and development efforts aimed at enhancing the hardware and performance of their balloons. Sushko noted that the contract was awarded without competitive bidding, reflecting the government's determination that WindBorne was uniquely capable of providing the global data collection required for these operations.

Further bolstering its capabilities, WindBorne CEO John Dean announced in August that the company had successfully raised a $37 million Series B funding round, dedicated to scaling its operations. The company also recently unveiled WeatherMesh 6, the latest iteration of its forecasting model. WindBorne asserts that WeatherMesh 6 delivers more accurate forecasts compared to the traditional model employed by the European Centre for Medium-Range Weather Forecasts (ECMWF), which relies on simulating the atmosphere using physical laws.

The advent of transformer model architectures and the release of a vast weather dataset by the ECMWF several years ago have propelled AI weather forecasting into one of the most productive applications of deep-learning technology. Historically, meteorologists depended on expensive supercomputers running complex equations to approximate atmospheric behavior. Now, end-to-end learning systems can process decades of weather data to generate predictions much more rapidly. Models developed by leading research labs and government agencies are increasingly demonstrating superior accuracy over conventional forecasts, although the field remains dynamic, with both traditional and AI-driven forecasts often necessary for a comprehensive understanding of weather patterns. Google DeepMind's advanced weather model, for instance, offers significant new capabilities, with its cyclone forecasting tools becoming central to how the U.S. government prepares for hurricane season.

In a groundbreaking achievement in August, WindBorne became the first organization to deploy an unpowered flight system — lacking an engine — to gather data directly within the eye of a typhoon. The company skillfully maneuvers its balloons by precisely adjusting their altitude, releasing air, or dropping sand used as ballast, enabling them to ride specific wind currents to their desired locations. Sushko highlighted the novelty of this feat, attributing its success to two key factors: having sophisticated wind and weather models for navigation, and possessing a sufficiently large constellation of balloons positioned in opportune locations. He emphasized that this marked one of the initial instances where a deep-learning model was leveraged for actual operational decisions in such a critical scenario.