Nvidia launches open AI weather models that beat Google DeepMind on 70+ forecast variables

Nvidia Releases Open-Source AI Weather Models That Outperform Google DeepMind

Nvidia announced a comprehensive suite of open-source AI models designed to transform weather forecasting from an expensive, supercomputer-dependent process into an accessible technology for governments, startups, and enterprises worldwide 1

1

. The company revealed the Earth-2 family of models at the American Meteorological Society meeting in Houston, timing the release amid a major winter storm that exposed inconsistencies in traditional forecasting methods 1

1

.

The flagship Earth-2 Medium Range model beats Google DeepMind's GenCast AI system on more than 70 weather variables, including temperature, pressure, wind, and humidity 1

1

3

3

. GenCast, released by Google in December 2024, was already significantly more accurate than existing models capable of generating forecasts up to 15 days out 1

1

. Mike Pritchard, director of climate simulation at Nvidia, described the approach as "a return to simplicity," moving away from hand-tailored architectures toward scalable transformer architectures 1

1

.

AI-Powered Weather Forecasting Slashes Computation Time and Costs

The Earth-2 suite includes three new models alongside two existing ones, creating what Nvidia calls the world's first fully open, accelerated set of models and tools for AI weather prediction 2

2

. Medium Range delivers 15-day high-accuracy forecasts powered by a new architecture called Atlas 3

3

. The Nowcasting model, powered by StormScope, uses generative AI to transform country-scale forecasts into kilometer-resolution predictions for zero to six hours, targeting severe storm predictions and hazardous weather 1

1

3

3

.

The Global Data Assimilation model, powered by HealDA, produces initial atmospheric condition snapshots from weather stations, balloons, and satellites at thousands of locations worldwide 3

3

. Traditionally, this data assimilation process consumed roughly 50% of total supercomputing loads for traditional weather forecasting, requiring hours on supercomputers 1

1

. The new model completes this work in minutes on GPUs instead 1

1

. The Israel Meteorological Service reported a 90% reduction in compute time at 2.5-kilometer resolution compared with running classic numerical weather prediction models on CPU clusters 2

2

.

Energy Companies and Financial Firms Deploy Earth-2 for Operational Decisions

Energy companies are among the earliest adopters of accelerated weather forecasting technology. TotalEnergies is evaluating Earth-2 Nowcasting to improve short-term risk awareness, with Emmanuel Le Borgne, climate and weather forecast product manager, calling it "a major step forward in how advanced weather intelligence can be operationalized at scale" 2

2

3

3

. GCL, one of China's largest solar material producers, is running Nvidia Earth-2 models operationally for its photovoltaic prediction system, achieving more accurate predictions at lower cost 2

2

.

The Weather Company is evaluating Nowcasting for localized severe-weather applications, while Southwest Power Pool, collaborating with Hitachi, is using Nowcasting and FourCastNet3 to improve wind forecasting for grid reliability 2

2

. The insurance industry stands to benefit significantly from the technology's speed advantage. Once trained, AI is 1,000 times faster than traditional methods, allowing insurance companies to run 10,000-member ensembles to understand extreme weather events like major floods or hurricanes 5

5

. S&P Global Energy is using CorrDiff to turn climate data into local insights for risk assessment, while global insurance group AXA is using FourCastNet to generate thousands of hypothetical hurricane scenarios 2

2

.

Open Models Aim to Democratize Weather Forecasting for Smaller Nations

Pritchard emphasized that the open-source AI models provide "fundamental building blocks used by everyone in the ecosystem -- national meteorological services, financial service firms, energy companies -- anyone who wants to build and refine weather forecasting models" 1

1

4

4

. The initiative seeks to democratize weather forecasting, which has historically been the domain of wealthier countries and large corporations with funds to pay for costly supercomputer time 1

1

.

Because Nowcasting is trained directly on globally available geostationary satellite observations rather than region-specific physics model outputs, its approach can be adapted anywhere on the planet with good satellite coverage 1

1

. This capability matters particularly for smaller countries where weather prediction is a national security issue. "For some users, it makes sense to subscribe to an enterprise centralized weather forecasting system. But for others like countries, sovereignty matters," Pritchard said. "Weather is a national security issue, and sovereignty and weather are inseparable" 1

1

.

Brightband, an AI weather tool provider and member of the Nvidia Inception program's Sustainable Futures initiative, is among the first to run Earth-2 Medium Range operationally, issuing real-world global forecasts daily 2

2

. Julian Green, cofounder and CEO of Brightband, noted that "the model being open source speeds up innovation, allowing easier comparison and improvements by other members of the weather enterprise" 2

2

3

3

. The full-featured open-model family will be licensed for commercial and noncommercial use and made available on GitHub and Hugging Face 3

3

.