How Nvidia’s new AI weather models may have predicted the storm long before it hit

DCM Editorial Summary: This story has been independently rewritten and summarised for DCM readers to highlight key developments relevant to the region. Original reporting by Tech Crunch, click this post to read the original article.

If you’ve been following the chaos around the latest U.S. winter storm and the confusing forecasts leading up to it, Nvidia’s timing couldn’t be better. The company just introduced its new Earth-2 suite of AI-powered weather forecasting models, including the standout Earth-2 Medium Range. Nvidia claims it outperforms Google DeepMind’s GenCast model on over 70 variables. That’s noteworthy, given GenCast already made headlines in late 2024 for its accuracy in medium-range forecasting, up to 15 days in advance.

These new models were revealed at the American Meteorological Society meeting in Houston. According to Nvidia’s climate simulation director, Mike Pritchard, the company is embracing simpler, more scalable AI architectures rather than specialized solutions. The Medium Range model is built on a new AI system called Atlas, designed to boost speed and precision in forecasting.

You’ll also find two other key tools in the Earth-2 suite. Nowcasting is built for short-term predictions – between zero and six hours ahead – and uses global satellite data to assist with rapid-response needs in severe weather. Because it’s based on satellite observations rather than region-specific inputs, the model works globally wherever satellite data is available. Meanwhile, the Global Data Assimilation model pulls from sources like weather stations and balloons to create real-time snapshots of weather conditions across thousands of locations, significantly cutting computing time from hours to minutes.

These join Nvidia’s earlier models like CorrDiff and FourCastNet3, which deliver high-resolution and variable-specific forecasts. By making these tools accessible via GPU processing, Nvidia aims to democratize advanced forecasting, allowing smaller nations and less-funded organizations to benefit from capabilities once limited to wealthy countries or enterprises with supercomputers.

Some applications are already in motion. Meteorologists in Israel and Taiwan are working with CorrDiff, while companies like The Weather Company and Total Energies are testing Nowcasting. For governments in particular, Nvidia emphasizes how crucial it is to have autonomous control over weather prediction systems — not just for safety, but for national security reasons.