NVIDIA Earth-2: Meteorological Supercomputers Go Retirement

Remember the times when the phrase "tomorrow it might rain" sounded like fortune-telling with coffee grounds, backed up by a week of calculations on a state supercomputer the size of a hangar? NVIDIA has decided it's time to end this era of elite forecasting. The company has presented Earth-2 — the world's first fully open technology stack for accelerated AI-powered weather and climate forecasting. This is not just another piece of software, but a fundamental shift in how humanity plans its life in conditions of changing climate.

Until now, weather prediction relied on pure physics and numerical methods that required colossal computing power. State meteorological services spent millions of dollars maintaining infrastructure to produce a three-day forecast. NVIDIA breaks this barrier by offering a family of models, including FourCastNet and CorrDiff, that work orders of magnitude faster than traditional methods. Now what used to take hours of supercomputer work, AI does in seconds on an ordinary server with graphics processors.

The CorrDiff model deserves special attention. It works like a "smart upscaler" for weather: it takes data from global models with low resolution and transforms them into ultra-sharp local forecasts with resolution down to two kilometers. This is critically important for predicting extreme phenomena such as typhoons or sudden floods, where an error of a couple of kilometers could cost hundreds of lives. Jensen Huang is essentially giving the industry tools that were previously available only to the chosen few.

Why is NVIDIA being so generous? The answer lies in the "razor and blade" strategy. By making software open and free, the company turns its H100 chips and future Blackwell into the only logical choice for climate startups and government agencies. While Google develops its GraphCast and Microsoft works on Aurora, NVIDIA is creating an entire ecosystem where every developer can fine-tune the model for their own needs. This is a classic move for market capture: become the foundation upon which everything else is built.

The transition from traditional physical models to neural networks (Neural Weather Prediction) is not just a trend, but a necessity. We have entered an era where climate changes faster than we manage to rewrite the formulas. AI allows us to adapt to these changes on the fly, learning from new data in real time. If previously a meteorologist was a mathematician with a logarithmic ruler, now he becomes an operator of the planet's digital twin.

The bottom line: NVIDIA Earth-2 makes climate analytics accessible to businesses of any scale. Will government meteorological services be able to restructure as quickly as the private sector will?