Google DeepMind and Google Cloud to provide European researchers up to 200 with frontier-model access and $3 million in compute credits

Google DeepMind and Google Cloud said they will back European scientific work by giving up to 200 researchers direct access to Google’s frontier AI models plus $3 million in compute credits. The announcement frames the effort as part of a broader push to expand high-performance computing capacity for scientists across Europe, where the company says researchers can sometimes face constraints obtaining large-scale compute.

Under the pledge, scientists will receive direct access to Google’s “leading scientific tools,” which the company lists as AlphaFold 3, WeatherNext, AlphaGenome, AlphaEvolve, and AlphaEarth Foundations. Google did not describe eligibility criteria, project selection timelines, or how the compute credits will be distributed beyond the aggregate amount and the stated cap of 200 researchers.

Google said the initiative is designed to support the European Union’s AI for Science work by making some of its frontier models available via the RAISE initiative. The company described RAISE as an effort intended to help bridge a compute gap that European researchers may encounter when trying to run advanced AI workloads.

The company linked the program to a goal of accelerating the pace of European innovation and strengthening the continent’s scientific competitiveness. In its description, Google positioned compute access and model availability as the practical inputs that can let research teams iterate faster on complex questions, rather than waiting longer for hardware resources or constrained allocations.

In the examples Google highlighted, researchers could use the tools to tackle extreme weather forecasting, mapping molecular structures, and other discovery tasks that depend on large-scale modeling. The announcement also implied that access to multiple model capabilities is intended to cover different domains of scientific research, from biology-oriented programs such as AlphaFold 3 to Earth and weather modeling represented by WeatherNext and AlphaEarth Foundations.

Google did not provide a breakdown of which specific frontier models will be available to all participants versus which tools may be offered by research area. It also did not disclose whether the compute credits correspond to particular workloads, usage limits, or a trial period, nor did it specify whether participants will publish results under the program’s terms.

For context, Google has previously presented several of the named models as specialized systems for scientific prediction and simulation. AlphaFold 3 is presented by Google as a tool for molecular structure prediction, while WeatherNext, AlphaGenome, AlphaEvolve, and AlphaEarth Foundations are positioned as capabilities aimed at forecasting or modeling across different scientific areas. In this latest pledge, Google is essentially bundling access to those systems and the compute needed to run them into a single research support effort.

The company’s focus on “direct access” and compute credits reflects a recurring challenge in AI for science: even when models are available, running large training or inference workloads can require substantial compute budgets. By offering credits and access to frontier tools, Google is trying to reduce friction between cutting-edge model availability and the operational realities of research labs.

One caveat is that Google’s announcement does not include measurable outcomes such as expected publications, patents, or research milestones. It also does not state how performance will be evaluated, what data governance or security requirements will apply to participating teams, or what terms will govern model usage and results sharing.

Looking ahead, researchers and policymakers will likely watch for further details on application and selection processes, the scope of model access under RAISE, and whether the program expands beyond the stated cap of up to 200 researchers. The impact will also depend on how quickly teams can turn compute access into experiments and how the initiative coordinates with existing European research programs and infrastructure.