Google.org gives $36 million to 17 AI-for-science teams, from fusion to storm warnings

What did Google.org announce?

Google.org, Google’s philanthropic arm, named the recipients of its Impact Challenge: AI for Science on 14 October 2026. Seventeen organisations will receive a total of $36 million, and each can choose to join a six-month Google.org Accelerator that provides engineering support, technical mentorship and access to Google infrastructure. Maggie Johnson, Google.org’s global head, wrote that the call drew applications from 30 countries.

The challenge opened on 18 February 2026 as a $30 million global open call, building on an earlier, smaller AI for Science fund. It asked for projects in health and life sciences, crisis resilience and environmental science, and applications closed on 17 April. The final pot is $6 million larger than first advertised; Google did not say why.

Google describes the goal in big terms, saying it wants to help researchers “unlock Nobel-level breakthroughs.” The selected projects are more concrete than that. Most pair a specific scientific bottleneck, such as forecasting battery wear or mapping how genetic variants affect RNA splicing, with a machine-learning method that could speed it up.

Who got funding for climate and environmental science?

Eight recipients work on climate resilience and environmental science. The most immediately practical is AfriClimate AI, which is studying the atmospheric precursors of deadly storms across Africa with the aim of extending flash-flood warning times from under three hours to more than six. More warning time would give communities longer to move to safety.

Two projects target fusion energy. The Max Planck Institute for Plasma Physics, working with TU Eindhoven, is using physics-informed AI to model the chaotic conditions inside fusion reactors and close the gap between simulation and real machines. The University of Texas at Austin is building neural-network models to speed up the design of compact fusion reactors.

Others focus on Earth systems and materials. Inria Chile is building a multimodal foundation model of the ocean, a digital twin of marine ecosystems. UC San Diego is developing an agentic pipeline that discovers interpretable climate physics from Earth observations, builds a climate model and decides where to collect new ocean measurements. MIT is training AI agents to interpret microscope images on their own to speed up automated materials discovery.

The University of Cambridge is combining physics foundation models with microscopy to forecast cathode degradation in sodium-ion batteries, a potential alternative to lithium-ion. The University of Washington is designing proteins from natural-language descriptions for sustainability uses such as breaking down plastic.

Who got funding for health and life sciences?

Nine recipients work in health and life sciences. Several aim to make medicine more precise for groups that research has under-served. The University of Cape Town is building a platform to map genetic variation in African drug-metabolising genes, to guide drug choices and improve safety. The Australian National University leads an international effort to end trial-and-error chemotherapy and make precision oncology more equitable.

Others are basic biology. The National University of Singapore is mapping how genetic variants regulate RNA splicing at population scale to predict disease effects. NYU Grossman School of Medicine is using AI to detect different types of DNA damage and uncover unknown causes of mutations. The University of Pittsburgh is comparing heart regeneration across species with a geometric AI model to find genetic switches for cardiac therapies.

Drug and protein design is the third cluster. The AITHYRA Research Institute, with EPFL, is studying how viruses manipulate cells and building an AI-designed protein library aimed at proteins that conventional drugs cannot target. The University of Toronto is mapping the design space for peptide-based drugs with a foundation model. The Allen Institute is designing DNA sequences that reach the brain cell types most vulnerable to neurodegenerative disease, and says it will openly share every model, sequence and result.

UC Berkeley’s project is about trust rather than discovery: it aims to turn black-box predictions from medical AI models into clear clinical biomarkers by extracting hypotheses that doctors can test.

What else Google offered European scientists the same day

Google paired the grants with a European announcement. In a post on the same day, Maureen Costello of Google Cloud said Google DeepMind and Google Cloud will give up to 200 European scientific researchers direct access to frontier models, together with $3 million in compute credits. The tools named are AlphaFold 3, WeatherNext, AlphaGenome, AlphaEvolve and AlphaEarth Foundations.

The access runs through RAISE, the Resource for AI Science in Europe. According to the European Commission, RAISE is a virtual research institute meant to pool compute, data, talent and funding for European researchers working on and with AI. It was set out in the EU’s AI in Science strategy in October 2025, launched in November 2025, and its pilot is funded by Horizon Europe with €33 million in the 2025 work programme and €107 million for 2026–27.

Google did not say how the 200 researchers will be chosen or how much access each gets. The tools themselves are not new: AI Lookout has covered the AlphaGenome Atlas of DNA variant effects and AlphaProtein-designed enzymes. What is new is a structured route for European labs to use them with paid compute.

Why tech companies are paying for AI science

Philanthropy and compute credits are now a standard part of how AI companies court scientists. In the United States, companies made pledges to the White House’s Genesis Mission for AI-driven science, and Chan Zuckerberg’s Biohub has put large sums into virtual biology. Google’s grants are small next to those, but they reach universities and nonprofits on six continents, including Africa, where research compute is scarce.

There is also a commercial logic. Researchers who build on Google’s models and cloud during an accelerator are likely to keep using them, and results in fields such as fusion, batteries or drug design make strong case studies. The European pledge landed the same day Google published an EU economic impact report arguing for a “predictable, supportive regulatory environment,” a reminder that science funding and policy goals travel together.

What is not disclosed

Google’s announcements leave several gaps:

• How the $36 million is split among the 17 organisations; no per-grant amounts were published.
• Why the pot grew from the $30 million advertised in February, and how many applications were received.
• The selection criteria and reviewers for both the Impact Challenge and the 200 European researcher places.
• Whether recipients must use Google models or cloud, and who owns the resulting models and data, apart from the Allen Institute’s pledge to share openly.
• Milestones or reporting dates for the funded projects.