Google DeepMind Science AI Models Open to 200 EU Researchers

What Google is offering

Google DeepMind and Google Cloud plan to give up to 200 European scientists access to a suite of the lab's scientific AI models, along with $3 million in cloud computing credits 15. The package centres on AlphaFold 3, the protein and molecular structure predictor, and WeatherNext, DeepMind's forecasting system 3. It also includes three newer tools: AlphaGenome for DNA analysis, AlphaEvolve, an agent that designs algorithms, and AlphaEarth Foundations, which builds maps of the planet from satellite data 12.

The programme will run through RAISE, the Resource for AI Science in Europe. The European Commission launched RAISE in November 2025 as a virtual institute, with €107 million in pilot funding 23. Most of that money goes toward training and attracting AI researchers 3. The announcement reportedly came on October 14, 2026, at an AI summit in Dublin 4. Reuters reported it as a Wednesday announcement 5.

Google presents the offer as a way to narrow the computing gap that European researchers often face. Maureen Costello, a Google Cloud vice president, wrote that the tools can help teams move faster on hard problems, from extreme-weather forecasting to molecular mapping 1. The company also says it wants to secure compute time for scientists at the EU's planned AI facilities 1.

The details that are still missing

Outlets report the logistics differently. Several say Google has not disclosed selection criteria, start dates, or how the credits will be divided among recipients 23. Other reports say the European Commission, not Google, will choose the participants under RAISE 45. Both accounts can be true at once: the Commission may run selection without a public rubric existing yet. Either way, applicants currently know little about how to get in.

The money is also modest when spread out. An even split across all 200 places would give each researcher about $15,000 in credits 4. That can pay for real experiments. It is not the scale of compute needed to train frontier models from scratch. In practice, this looks like a programme for running and fine-tuning Google's existing tools, not for building rival systems.

AlphaFold 3 access is not entirely new either. DeepMind released the model in May 2024 with its drug-discovery sister company Isomorphic Labs. Academics have been able to download its code and weights for non-commercial use since November 2024 1. For AlphaFold specifically, the added value is mainly managed compute and integration, not first-time access.

The AlphaFold team context

The announcement follows a period of upheaval for the group that made DeepMind's scientific reputation. In July, the company broke up the AlphaFold team, whose work earned a Nobel Prize in 2024, and moved most of its members to other projects 3. John Jumper, one of the Nobel laureates, left about a month before that 2.

None of the reports link the two events directly. The timing still invites a question. DeepMind is promoting AlphaFold 3 to European scientists just months after dissolving the team that built it. One reading is that the model has matured into a product. On that view, it needs maintenance and distribution more than a dedicated research group, and broad access through cloud partnerships is the natural next step.

A less generous reading is that Google is turning a celebrated research asset into a channel for cloud adoption. The $3 million arrives as compute credits on Google Cloud. Researchers who build workflows around these models on that infrastructure may stay there after the credits run out.

Why it matters

The deal matters most as a signal about how AI-driven science is being distributed. Europe has put public money into RAISE to strengthen its own AI research base 23. Through this programme, part of that ecosystem will run on tools and infrastructure owned by a US company. That arrangement is pragmatic, because European labs get capable models quickly. It also highlights a dependency that EU policymakers have talked about reducing.

On balance, this looks like a useful but narrow programme. Two hundred researchers getting hands-on access to AlphaGenome, AlphaEvolve and WeatherNext could produce interesting results, especially in climate and genomics. The newer models may offer more to European researchers than AlphaFold 3, which is already available to academics.

The bigger story is strategic. DeepMind appears to be moving its scientific flagships from in-house research projects to platforms it distributes through partnerships. The breakup of the AlphaFold team fits that shift, and this European offer is one of its first public examples. Whether the scheme succeeds will depend on details Google and the Commission have not yet released: who is selected, when access begins, and what happens after the credits are spent.