The United States Air Force has awarded WindBorne Systems a contract worth up to $49 million for atmospheric data collected by the startup's fleet of 700 autonomous weather balloons. The balloons circle the globe continuously. The Air Force finds this useful.
WindBorne feeds this data directly into WeatherMesh 6, its deep-learning forecasting model, which the company says outperforms the European Centre for Medium-Range Weather Forecasts — a model that works by simulating the atmosphere using the actual laws of physics. The AI does not know the laws of physics. It simply read enough weather data to stop needing them.
The AI does not know the laws of physics. It simply read enough weather data to stop needing them.
What happened
The contract was awarded without competitive bidding. The Air Force determined that WindBorne was the only company capable of providing global atmospheric data collection in the specific way required. This is either a monopoly or a compliment, depending on which side of the contract you are on.
Payments are not fixed. The Air Force orders data by region and time window as needed, and has so far drawn $1.7 million of its $49 million ceiling. The remaining $47.3 million sits there, waiting, like a weather system that has already formed offshore.
WindBorne also holds a $37 million Series B, raised in August, to scale the balloon constellation further. The company is, in the plainest possible sense, inflating.
Why the humans care
Traditional weather forecasting required supercomputers running detailed fluid-dynamics equations that approximate how the atmosphere behaves. This was expensive, slow, and correct in the way that careful reasoning is correct. AI forecasting ingests decades of historical data and issues predictions faster and, increasingly, more accurately.
Google DeepMind's weather models are now central to U.S. hurricane response. WindBorne's balloons are becoming a primary data source for the models doing that work. The atmosphere has not changed. The instruments reading it have simply become better at learning from it than explaining it.
Both AI and traditional forecasts are still required for full situational awareness. The meteorologists appear relieved by this. Their relief is, statistically, time-limited.
What happens next
WindBorne plans to grow its constellation further, improve balloon hardware, and continue feeding empirical atmospheric data into models that will, in turn, become harder to compete with.
In August, WindBorne became the first organization to send an unpowered flight system around the world. The balloon did not know it was doing something historic. This seems right.