Google.org awards $36 million to 17 teams pursuing AI-powered science

Google.org backs UT Austin team to speed fusion reactor design with AI

A University of Texas at Austin team has won three years of Google.org support to tackle a costly step in fusion research: testing reactor designs on computers. Led by physicist Joshua Burby, the project will develop AI models intended to make the search for compact stellarator reactors faster and more affordable, the university said October 16.

The award comes through the Google.org Impact Challenge: AI for Science. UT Austin is one of 17 organizations sharing $36 million, selected from more than 3,000 applicants across 30 countries. The university did not disclose its team’s individual award, so the program-wide total should not be read as the budget for this fusion project.

A stellarator uses precisely shaped magnetic fields to hold the superheated plasma that fuels fusion in place. Its twisted, doughnut-like shape is asymmetrical, making each candidate design a complicated computational problem. Researchers must work out how heat and particles move inside the reactor while adjusting the surrounding magnetic field to improve the design.

The existing approach relies on detailed physics simulations, run repeatedly as researchers refine those shapes. Those calculations are slow and expensive. The team’s proposed alternative is to build neural-network surrogate models: AI stand-ins intended to take over most of that repeated calculation work, rather than require the full simulation effort for every design iteration.

The intended payoff is a wider search. With less time and money spent evaluating each option, researchers could examine more configurations and identify economically attractive reactor candidates. UT Austin says those designs have the potential to lower future power-plant construction costs. That is the project’s hoped-for consequence, not a cost reduction demonstrated by the award.

Burby says the team aims to reach stellarator configurations that existing design tools cannot currently access, while ensuring model consistency. The work builds on the Oden Institute’s research in model-constrained scientific machine learning. Speed is therefore only one part of the stated research agenda; the team also identifies accuracy and physical consistency as requirements.

This award gives us an opportunity to push scientific machine learning into a difficult multiphysics design problem where accuracy, generalization, uncertainty, and physical consistency all matter.

The collaboration brings together UT Austin, the University of Washington and New York University. Burby, an assistant professor of physics, is the principal investigator. The project divides responsibility for machine learning, high-performance computing and design optimization among researchers with different specialties:

Tan Bui-Thanh, a UT Austin professor of aerospace engineering and engineering mechanics, leads the machine-learning work.

David Hatch, a UT Austin research associate professor of physics, leads high-performance computing.

Alan Kaptanoglu, a University of Washington assistant professor of aeronautics and astronautics, leads optimization.

Project funds will support three new postdoctoral researchers at UT Austin and one at the University of Washington. The wider collaboration also includes researchers from the Dutch Institute for Fundamental Energy Research, Type One Energy and ExoFusion. Alongside the academic team, Google employees Nova Fallen and Edouard Gahou will provide research and engineering support.

The university’s detailed project announcement says all machine-learning models and agentic workflows produced by the project will be available in open-source repositories. That commitment covers both the models and the AI-driven workflows developed around them, rather than keeping the research software solely within the participating institutions.

UC San Diego receives $2.35 million from Google.org to link climate AI with ocean measurements

UC San Diego researchers have received $2.35 million from Google.org for climate AI that reaches beyond computer models. Announced October 14, 2026, the award supports a team led by Duncan Watson-Parris that plans to improve climate modeling and use the work to guide new ocean observations, including measurements of small-scale mixing.

The project starts with a problem Watson-Parris identifies in climate modeling: powerful models still struggle to represent some important physical processes. His team wants to use AI to learn better descriptions of those processes from data. The aim is not simply to produce another prediction, but to improve the physics represented inside the model.

Google.org’s announcement describes a connected pipeline using AI agents, with three intended tasks. Google lays out the planned progression this way:

Learn climate physics from Earth observations. The first task is to discover understandable physical relationships in measurements of the planet, giving the modeling effort a foundation in observed data.

Build a new climate model. Google describes this as a next-generation model, placing model development between the effort to learn physics and the decisions about gathering more observations.

Decide where to collect new ocean observations.

This support gives us an opportunity to use AI to help learn improved physics from data, identify where new observations could have the greatest value and then go into the ocean and collect those measurements.

The ocean component gives that plan a specific measurement target. The team intends to use data collected by specially adapted Argo floats that measure small-scale ocean mixing.

Rather than relying only on existing datasets, the collaborators plan to let insights from the modeling effort inform future scientific measurement efforts. Better climate models and more useful observation choices remain research goals, not improvements demonstrated by the grant announcement.

Watson-Parris is an assistant professor jointly appointed at UC San Diego’s Halıcıoğlu School of Data Science and Scripps Institution of Oceanography. An atmospheric physicist, he studies interactions among aerosols, clouds and climate, alongside machine-learning approaches to improving climate models across scales. His research combines physical models with data-driven methods to improve understanding of the climate system and its projections.