A fleet of balloons with no engines, drifting entirely on the wind, has just landed a big order from the US Air Force. WindBorne Systems, a company that makes autonomous weather balloons, announced it has won a new contract from the Air Force, which will purchase data collected by its global balloon fleet, with a total value of up to $49 million. For a startup whose business is "peering into the atmosphere with balloons," this is the heaviest deal it has ever been part of.
WindBorne currently has about 700 weather balloons floating around the Earth. They continuously read data from atmospheric sensors, selling some of it to paying customers while feeding the rest back into the company's own AI weather forecasting model. Founded in 2015 in Redwood City, California, the company was one of the earliest organizations to feed real observation data directly into deep learning weather models. It has long dealt with the US government—the National Weather Service, the Air Force, and the Navy are all old acquaintances—but this Air Force contract truly pushes the collaboration to a new level of scale.
The money is not being spent recklessly. The contract is capped at $49 million, but the Air Force "pays only for what it uses"—placing orders for specific regions and time periods only when data is needed. So far, only $1.7 million in orders have actually been placed, and company executives expect more to follow. Co-founder and CTO Andrey Sushko described the contract as "primarily a vehicle for supporting the constellation's continued expansion, as well as funding R&D to keep improving the balloons' hardware and performance." He specifically noted that the contract was awarded without competitive bidding, because the government determined that "in collecting global data, we are the only company able to provide this capability in a way that meets its operational needs."
WindBorne's confidence comes from solid technical groundwork. CEO John Dean disclosed in August that the company had just completed a $37 million Series B funding round to keep fueling its business expansion; and the company claims its WeatherMesh6 forecasting model, released in early summer, is already more accurate than the traditional model of the European Centre for Medium-Range Weather Forecasts (ECMWF), which simulates the atmosphere using physical laws. Since the transformer architecture emerged and ECMWF released massive meteorological datasets, AI weather forecasting has become one of deep learning's most powerful applications—whereas meteorologists once relied on expensive supercomputers to solve complex equations approximating atmospheric behavior, end-to-end learning models can now ingest decades of data and produce predictions faster. Models from frontier labs and government agencies are indeed getting more accurate, but the two types of forecasting still need to complement each other for people to truly understand the weather; for example, Google DeepMind's latest weather model, with its hurricane forecasting tools, has become increasingly critical to government responses during storm season.
The most validating moment for WindBorne came in August this year, when it became the first organization to send an unpowered flight system into a typhoon's eye to collect data. These balloons have no engines; they rely entirely on adjusting their flight altitude, releasing gas, or dropping sand used as ballast to ride the wind to where they need to go. Sushko called this a remarkably novel move: "There are two prerequisites: first, you need good wind and weather models and know how to work with them; second, the constellation must be large enough for the balloons to happen to be in the right position." He added a weighty remark—this was one of the first times deep learning models were used for genuine operational decision-making. In other words, these wind-drifting balloons are no longer just scientific playthings; they have begun making real, practical judgments for the military and weather agencies.