NVIDIA Earth-2: Open Source AI Predicts Weather 1000x Faster
NVIDIA launches Earth-2, a family of open source AI models for weather forecasting. Predictions up to 15 days with 90% less compute time than traditional methods.
While a winter storm battered the U.S. East Coast, NVIDIA unveiled at the American Meteorological Society conference its most ambitious climate project: Earth-2, the first complete family of open source AI models for weather forecasting.
What is Earth-2?
Earth-2 is a complete accelerated AI software stack for climate prediction:
| Component | Description |
|---|---|
| Models | 3 specialized open source models |
| Libraries | Tools to train and run |
| Frameworks | Integration with existing workflows |
| Acceleration | Optimized for NVIDIA GPUs |
The 3 Earth-2 Models
1. Earth-2 Medium Range (Atlas)
Weather predictions up to 15 days in advance:
- Over 70 climate variables (temperature, pressure, wind, humidity)
- Superior to Google DeepMind GenCast in accuracy
- Global resolution with regional detail
2. Earth-2 Nowcasting (StormScope)
Storm forecasts from 0 to 6 hours with kilometer resolution:
- Storm development prediction
- Cloud system evolution
- Precipitation patterns
- Trained on satellite and radar data
3. Earth-2 Global Data Assimilation (HealDA)
Generates initial atmospheric conditions in seconds (vs hours on supercomputers):
| Traditional Method | Earth-2 HealDA |
|---|---|
| Hours of compute | Seconds |
| CPU supercomputers | Standard GPUs |
| High operational cost | Reduced cost |
Available later in 2026
Production Results
Organizations are already using Earth-2:
Israel Meteorological Service
“NVIDIA Earth-2 models give us a 90% reduction in compute time at 2.5-kilometer resolution compared with running a classic numerical weather prediction model without AI on a CPU cluster.” — Amir Givati, Director
Industry Adoption
| Sector | Organizations |
|---|---|
| Meteorology | Israel Meteorological Service, Taiwan CWA, US National Weather Service, The Weather Company |
| Energy | TotalEnergies, Eni, GCL, Southwest Powerpool |
| Finance/Risk | AXA, JBA Risk Management, S&P Global Energy |
| Startups | Brightband |
Speed: 1000x Faster
NVIDIA’s main claim: AI-powered weather forecasting that’s approximately 1000 times faster than traditional physics-based simulations.
Availability
| Model | Status | Access |
|---|---|---|
| Medium Range | Available | Hugging Face, GitHub, Earth2Studio |
| Nowcasting | Available | Hugging Face, GitHub, Earth2Studio |
| Data Assimilation | Q3-Q4 2026 | TBA |
Business Implications
Immediate Opportunities
- Renewable energy: Solar/wind generation prediction
- Agriculture: Planting and harvest planning
- Logistics: Weather-based route optimization
- Insurance: More accurate climate risk models
- Retail: Weather-based demand forecasting
Does your business depend on accurate weather forecasts? Contact us to evaluate how Earth-2 can integrate into your operations.
Sources
Have a project in mind?
Let's talk about how we can help you achieve your technology goals.
Schedule a free consultation