Nvidia Rolls Out Open-Source AI Weather Models as Federal Funding Wanes

Nvidia has announced a suite of open-source AI weather forecasting systems, joining other Big Tech players hoping to establish themselves in the space as federal funding evaporates.

California farmers, insurers and meteorologists alike stand to gain from adding AI to their weather-forecasting toolboxes.

At the American Meteorological Society’s annual meeting in Houston, Nvidia unveiled a new NVIDIA Earth-2 “family” of open models, libraries and frameworks for weather and climate AI, offering what it called “the world’s first fully open, accelerated weather AI software stack.”

The Santa Clara-based chipmaker described the system as “complete” for nowcasting and medium-range predictions that previously took hours on high-performance computing clusters.

Nvidia said the tools represent the first time AI has surpassed traditional, physics-based weather prediction models in short-term precipitation forecasting. The company added that developers across industries are already using Earth-2 to predict weather and “harness actionable insights.”

It’s a shot across the bow at other private AI developers, including Alphabet’s Google, Microsoft and Huawei Technologies.

Private-sector AI tools like Nvidia’s are welcome additions — not replacements — in a rapidly changing world, according to climate scientist Daniel Swain of UC Agriculture and Natural Resources and UCLA.

Swain said he is less concerned about the hallucinations that plague public-facing large language models than about AI weather modeling’s still unproven ability to predict edge cases based on historical data.

“Sometimes when it matters most — the very most extreme events that might be at the edge or outside of what we’ve seen historically — is precisely when we need the most accurate weather forecast,” Swain said. “We might not be there yet.” He added that the technology is rapidly advancing.

“There are real gains, in terms of scientific understanding as well as in prediction, and there’s need for continued caution,” said Noah Diffenbaugh, a professor and senior fellow at Stanford University’s Doerr School of Sustainability. But he struck a more cautionary note. “Other AI applications can produce inaccurate results, can produce results that are not grounded in reality. That’s a risk with these systems as well.”

Private developers trained their AI on a corpus of data that was largely publicly funded. While that bolsters the models’ credibility with scientists, it also raises troubling questions.

For one, private developers are, by definition, concerned with profit — eventually, if not immediately. There is no guarantee they will not begin charging for access to their models.

“Within a university context, we have no profit motivation at all,” Diffenbaugh said. “We’re trying to understand how the world works. And we’re doing that within our time scale, a much longer time scale (than private developers). And I think the benefit that we can bring in our work is that we’re doing that work in the context of this rigorous, patient scientific evaluation.”

The primary question for Swain is whether optimism about end-to-end AI models could be used by Trump administration officials to justify ceding data collection and weather modeling entirely to the private sector, even as global warming dramatically alters the climate system, particularly in California, with its complex interplay of atmospheric rivers, marine layers, Sierra snowpack, wind patterns and wildfire risk.

“Not only are we not there yet, not only do I think we won’t be there anytime soon, I’m not sure that we will ever get to that point,” Swain said. “It’s almost a category error to assume that the success of AI-based predictive modeling means that it’s just going to completely replace that whole pipeline. That’s just fundamentally divorced from the reality of the world we live in today, and very likely to be divorced from the reality of the world that we’re going to be living in for the foreseeable future.”