Nvidia AI Model Accurately Predicted Storm Weeks in Advance

As the winter storm battered much of the United States, weather forecasts for certain regions showed significant discrepancies, with snowfall predictions varying widely.

The timing of Nvidia's new Earth-2 weather forecasting models couldn't be more perfect. Or perhaps, given the claimed accuracy of these models, the company had advance insight into the weather conditions?

These new AI models are designed to deliver faster and more precise weather forecasts. Nvidia states that one specific model, Earth-2 Medium Range, outperforms Google DeepMind's AI weather model, GenCast, across more than 70 variables. GenCast, released by Google in December 2024, had already demonstrated superior accuracy compared to existing weather models capable of generating forecasts up to 15 days in advance.

Nvidia introduced these new tools on Monday at the American Meteorological Society meeting in Houston.

"From a philosophical and scientific standpoint, this represents a return to simplicity," Mike Pritchard, Nvidia's director of climate simulation, explained to reporters before the meeting. "We're transitioning away from specialized, handcrafted AI architectures and embracing the future of straightforward, scalable transformer-based systems."

Conventional weather forecasting primarily relies on physics-based simulations of real-world observations. AI models are a more recent development in this field. The Earth-2 Medium Range model is built on a new Nvidia architecture called Atlas, with further details scheduled for release on Monday.

In addition to Medium Range, Nvidia's Earth-2 suite includes a Nowcasting model and a Global Data Assimilation model.

Disrupt 2026 Tickets: One-time offer

Tickets are now available! Secure savings of up to $680 during this limited-time offer, and be among the first 500 registrants to receive 50% off your +1 pass. TechCrunch Disrupt features industry leaders from Google Cloud, Netflix, Microsoft, Box, a16z, Hugging Face, and many others across 250+ sessions focused on driving growth and enhancing your competitive advantage. Engage with hundreds of innovative startups and participate in curated networking opportunities that facilitate deals, provide valuable insights, and spark inspiration.

Disrupt 2026 Tickets: One-time offer

Tickets are now available! Secure savings of up to $680 during this limited-time offer, and be among the first 500 registrants to receive 50% off your +1 pass. TechCrunch Disrupt features industry leaders from Google Cloud, Netflix, Microsoft, Box, a16z, Hugging Face, and many others across 250+ sessions focused on driving growth and enhancing your competitive advantage. Engage with hundreds of innovative startups and participate in curated networking opportunities that facilitate deals, provide valuable insights, and spark inspiration.

Nowcasting generates short-term weather predictions from zero to six hours ahead, helping meteorologists assess the potential impacts of storms and other hazardous weather conditions.

"Since this model is trained directly on globally available geostationary satellite data rather than region-specific physics model outputs, Nowcasting can be adapted to any location with adequate satellite coverage," Pritchard noted. This capability should assist state governments and smaller nations in understanding how severe weather systems could affect their regions.

The Global Data Assimilation model processes data from sources including weather stations and balloons to create continuous snapshots of weather conditions at thousands of global locations. These snapshots serve as starting points for weather models to generate their forecasts.

Traditionally, producing these snapshots demanded enormous computing resources before forecasting could commence. "This process typically consumes about 50% of the total supercomputing capacity used in traditional weather forecasting," Pritchard explained. "Our model can accomplish this in minutes using GPUs instead of the hours required by supercomputers."

These three new models join two existing ones: CorrDiff, which uses coarse-grained forecasts to rapidly generate high-resolution predictions, and FourCastNet 3, which models individual weather variables such as temperature, wind, and humidity.

According to Pritchard, these new models should democratize access to advanced weather forecasting tools, which have historically been available mainly to wealthier nations and large corporations that can afford expensive supercomputer time.

"These tools provide fundamental building blocks for everyone in the ecosystem—national meteorological services, financial firms, energy companies—any organization interested in developing and refining weather forecasting models," Pritchard stated. Some tools are already operational, with meteorologists in Israel and Taiwan using Earth-2 CorrDiff, while The Weather Company and Total Energies are evaluating Nowcasting.

"While some users may prefer subscribing to centralized enterprise weather forecasting systems, others, particularly countries, prioritize sovereignty," Pritchard emphasized. "Weather represents a national security concern, making sovereignty and weather forecasting inherently interconnected."