This AI weather startup is out-forecasting government agencies
WindBorne Systems says its new WeatherMesh 6 model is more accurate than ECMWF forecasts on several variables, thanks to better data ingestion and balloon data.
Intelligence analysis by GPT-5.4 Mini
WindBorne Systems says its latest AI weather model is now beating the European governments' ECMWF on several forecast measures, aided by a large balloon data network. The company argues that owning both the sensors and the model gives it an edge in weather AI.
WindBorne is a company that launches weather balloons and uses the information they collect to help a computer predict the weather. The company says its newest system is better than some big government weather tools.
Think of it like a map that keeps getting fresh updates. If a team gets lots of new clues from balloons all the time, its weather guesses can become sharper than a team that only checks in less often.
The company is also trying to make sure its balloons stay safer for airplanes. It already sells its data to weather and military groups, and now it wants its computer forecast to be good enough to stand on its own.
Analysis
What changed
WindBorne Systems released WeatherMesh 6, the latest version of its AI weather model, and says it is now more accurate than the European Centre for Medium-Range Weather Forecasting on several variables. The company says the new model also produces forecasts every hour instead of every six hours, with 3 km resolution across Europe and the continental US, where its data quality is strongest.
Why WindBorne thinks it is winning
The company was founded by Stanford students in 2019 and originally focused on building better weather balloons and selling the data. After deep learning weather models became viable in 2022, WindBorne shifted to building its own model as well. Its leaders say the key advantage is not just model architecture but the data pipeline: roughly 400 balloons are in flight at any given time from 15 launch sites, and the company says improvements came from feeding those readings directly into the model.
CEO John Dean argues that an AI weather business without a data advantage is weak, and the company’s head of AI, Joan Creus-Costa, says direct ingestion of balloon data and other sources is the reason WeatherMesh improved. Dean also says the system was reworked for about a year to stay stable while using this approach.
How this fits into the wider market
The article frames ECMWF as the benchmark for accurate forecasting today, but notes that AI weather models are improving quickly. It also says AI forecasting is already being used by major government agencies and that researchers are working to fold it into public forecasting systems.
WindBorne has raised $25 million in venture funding and reported an $85 million valuation in 2024. It sells balloon data to NOAA and to the U.S. Air Force and Navy, and it also sells forecasts to investors and commodity traders. Even so, Dean says the company is prioritizing model and data infrastructure over building a larger consumer SaaS business.
The article also notes a safety scare last year, when a United Airlines jet hit one of WindBorne’s balloons, causing minor damage but no injuries. The company has since added ADS-B transponders to help reduce crash risk.
Key points
- WindBorne says WeatherMesh 6 is more accurate than ECMWF forecasts on several variables.
- The model now produces forecasts hourly and reaches 3 km resolution in Europe and the continental US.
- WindBorne says its edge comes from combining balloon data collection with model building.
- The company has about 400 balloons aloft at a time from 15 launch sites.
- WindBorne sells data to NOAA and the U.S. Air Force and Navy, and also sells forecasts to investors and traders.
If WindBorne's results hold up, its balloon network could help produce faster and more accurate forecasts than older systems in some regions. That could improve public weather services and make its data more useful for agencies, investors, and commodity traders.
The article shows that the system still depends on careful data handling and a year of tuning, so performance may not transfer cleanly everywhere. It also highlights aviation risk, since one of the company's balloons already caused a minor incident with a commercial plane.
