On June 1, WindBorne Systems released the sixth version of its AI weather forecasting tool, WeatherMesh, claiming it delivers more frequent and accurate predictions than the world-leading system developed by the European Centre for Medium-Range Weather Forecasts (ECMWF), the European intergovernmental organization that meteorologists consider the benchmark for accurate weather prediction.
One way to understand the accuracy claim: WeatherMesh-6 is, according to WindBorne chief product officer Kai Marshland, "as accurate five days out as a traditional forecast is the day before," particularly on surface temperature measurements. The model produces a forecast every hour, compared to every six hours for traditional physics-based models, and its resolution is now down to 3 km in the continental U.S.
WindBorne was founded by a group of Stanford students in 2019 and began by building a better weather balloon, with the idea of selling weather data. When deep learning forecasting models arrived in 2022, the team decided to build their own forecasting model rather than just supply the data.
Balloons as a moat
The company now has about 400 balloons in flight gathering sensor readings at any given time, launched from 15 sites around the globe. The advances in WeatherMesh-6 come from improvements in how that balloon data is fed directly into the models.
CEO John Dean told TechCrunch: "I don't understand, personally, the business model of being [an] AI based weather company without a dataset advantage." It is a pointed observation given how many AI startups compete on borrowed infrastructure. WindBorne collects its own.
The ECMWF's edge has historically come from its skill at "data assimilation" — the work of turning disparate sensor readings into a comprehensive, machine-readable picture of the world. For now, most AI weather models still depend on datasets produced by the ECMWF and the U.S. National Oceanic and Atmospheric Administration. WindBorne is trying to cut that dependency by feeding balloon data directly into its model rather than routing through a government intermediary.
WindBorne has raised $25 million in venture funding, with a reported valuation of $85 million in 2024. The company sells its balloon data to NOAA, where it is used in the American weather forecasting enterprise, and to the U.S. Air Force and Navy, and sells forecasts to investors and commodity traders.
The agent question
Traditional weather forecasts are generated by complex physics models that require expensive supercomputers and long run times. AI models — built by startups and labs including Google DeepMind — tend to move faster but currently lack the resolution or long-horizon accuracy of physics-based systems. WeatherMesh-6's claimed results, if independently validated at scale, would represent a meaningful shift in that dynamic.
The business model question is also changing. Dean said he is not trying to invest a large team in building a SaaS product if the way people want consumer information two years from now is through an agent. For a small business operator — a farmer managing crop risk, a logistics operator routing around severe weather, or a small commodities trader — that framing matters. If forecast data becomes something an AI agent pulls and acts on without a human opening a dashboard, the companies that own the underlying data pipelines will have leverage that pure-software forecast vendors will not.
Weather AI is improving rapidly and is already being used at major government agencies around the world. WindBorne is one of a small number of startups trying to own both ends of the problem — the sensors and the model — rather than competing on architecture alone. With 400 balloons up and a model that now updates hourly at 3 km resolution, the gap between what a $25 million company can produce and what a European intergovernmental body can produce got measurably smaller on June 1.
