WindBorne’s WeatherMesh-6 Claims Major AI Forecasting Gains

WindBorne Systems has released a new artificial intelligence weather model that it says can provide more accurate forecasts than some of the world’s leading government-backed systems. The bigger question is whether the startup can turn that technical lead into a profitable forecasting business.

Released on June 1, WeatherMesh-6 combines WindBorne’s AI model with observations collected by its own network of long-duration weather balloons. The company says the system can update global forecasts every hour, deliver 15-day ensemble predictions and produce higher-resolution forecasts for the continental United States every 15 minutes.

WeatherMesh-6 Claims a Major Accuracy Gain

WindBorne says WeatherMesh-6’s global model operates at approximately 25-kilometre resolution and uses 128 ensemble members to represent different possible weather outcomes.

In company-run evaluations covering July 2025 through March 2026, WeatherMesh-6 recorded up to 38% lower ensemble-mean error than the European Centre for Medium-Range Weather Forecasts’ physics-based IFS model. It also reported up to 32% lower error than ECMWF’s AIFS artificial intelligence system.

For two-metre surface temperature, WindBorne claims its forecast produced four and a half days in advance was about as accurate as an IFS forecast produced one day ahead. However, these results have been published primarily by WindBorne and have not yet received the same level of independent scrutiny as established forecasting systems.

The company also introduced a regional model covering the continental United States at approximately 2.5-kilometre resolution. WindBorne says it outperformed NOAA’s High-Resolution Rapid Refresh model for temperature and wind speed after the initial forecast hour. It updates every 15 minutes and currently extends roughly four days ahead.

WindBorne’s Balloons Provide Its Real Advantage

WindBorne’s strongest differentiator may not be WeatherMesh-6’s model architecture but the data being fed into it.

The company reportedly has around 400 autonomous balloons operating at any given time from 15 launch locations worldwide. Its balloons can remain airborne for weeks, change altitude and collect repeated atmospheric profiles instead of completing the single ascent associated with conventional radiosondes.

WeatherMesh-6 assimilates observations from WindBorne’s balloons alongside satellites, radar systems and surface-monitoring networks. The model includes 11 observation types and generates a revised forecast after incorporating newly available readings each hour.

CEO John Dean told TechCrunch that he did not understand the business model of running an AI weather company without a data advantage. WindBorne began in 2019 as a weather-balloon company but expanded into AI forecasting after recognising that it could capture more value by selling predictions instead of supplying observations alone.

Government Contracts Have Provided Early Revenue

WindBorne raised $15 million in Series A funding in 2024 in a round led by Khosla Ventures, with participation from Footwork VC, Pear VC and Convective Capital.

At the time, the company said it had already generated millions of dollars through government partnerships, research agreements and data-as-a-service contracts. It did not disclose an exact revenue figure or whether the business was profitable.

WindBorne has worked with NOAA and the Defense Innovation Unit and has received awards involving the US Navy and Air Force. A Navy contract focused on subseasonal and tropical-cyclone forecasting, while Air Force work included ensemble forecasting, higher-resolution models and edge computing for weather prediction. WindBorne said NOAA had been an operational customer since early 2023.

WeatherMesh Is Now an API Product

WeatherMesh-6 is available through a REST API, allowing businesses to integrate its forecasts into internal applications rather than relying on a consumer-facing weather service.

The global product includes variables covering temperature, wind, precipitation, soil moisture, cloud cover and solar radiation. WindBorne is positioning these outputs for industries including energy, agriculture and aviation.

Its documentation also includes heating-degree and cooling-degree calculations, tropical-cyclone data, point forecasts and gridded forecast products. Those services could make the model useful to utilities forecasting electricity demand, renewable-energy operators estimating wind and solar output, commodity traders and companies managing weather-related operational risks.

Accuracy Alone Will Not Guarantee Profit

WindBorne has established several elements needed to build a defensible weather-technology company: proprietary hardware, a growing global dataset, an operational AI model, government customers and an API through which commercial clients can access its forecasts.

What remains unclear is whether WeatherMesh can attract enough private-sector customers to support the expense of manufacturing, launching and operating a global balloon network.

Its benchmark lead also needs to hold up across extreme events, different regions and independent evaluations. Weather forecasts used for energy trading, aviation or disaster planning must remain reliable under the unusual conditions where errors can become most expensive.

WindBorne is therefore selling more than a faster forecast. Its commercial pitch is that owning both the atmospheric data and the AI model will allow it to deliver predictions that competitors relying on public datasets cannot easily reproduce.

WeatherMesh-6 provides stronger evidence that the technology can work. Whether WindBorne becomes lucrative will depend on how effectively it converts that forecast advantage into recurring contracts, API usage and industry-specific products.