NVIDIA Earth-2 Open Models Put Weather Forecasts in More Hands

Weather forecasting has long been a closed shop. National meteorological agencies run supercomputers worth hundreds of millions of dollars, and the models they produce are rarely shared. At GTC Berlin, NVIDIA and a cluster of European partners demonstrated a different approach: Earth-2, a family of open models for weather and climate prediction, now being deployed for real-world forecasting and climate-risk work across the continent.

The pitch is straightforward. Open models lower the barrier to entry, letting utilities, insurers, logistics firms and regional governments run their own forecasts rather than waiting on a national agency’s next update. NVIDIA’s accelerated computing stack supplies the horsepower, while the open-weight models supply the physics. The company says the same pattern is spreading beyond weather: at the Berlin event, banks, payment companies and financial data providers showed production-scale intelligence built on NVIDIA hardware and open models.

For global readers, the significance is less about any single forecast and more about who gets to make one. Weather and climate risk are increasingly priced into insurance, agriculture and energy trading. If open models make regional modelling cheap enough, smaller countries and private firms can build bespoke predictions for their own coastlines, harvests and power grids. That could improve resilience in places that cannot afford a supercomputer.

There are caveats. Open models still need data, and high-quality observational data is unevenly distributed. A model trained on sparse readings from one region may not transfer cleanly to another. NVIDIA’s partners are presenting production deployments, not peer-reviewed benchmarks, and the gap between a demo and a dependable operational forecast is wide.

The broader signal is the direction of the AI stack. NVIDIA is best known for the chips that train large language models, but Earth-2 is a bet that scientific simulation will be a durable second market. Open weights help here: they let researchers inspect and adapt a model, which matters when a wrong forecast has real consequences. The same logic is driving adoption in finance, where explainability and local control are prized.

What to watch next is whether national weather services treat these open models as complements or competitors. If a few large agencies adopt them openly, regional and private forecasters will follow quickly. If they resist, the technology may spread first through insurance and energy, where the appetite for granular risk data is already strong. Either way, the monopoly on forecasting is loosening.