Europe asked companies to stock the RAISE pantry. Researchers need the keys

THE SIGNAL IN ONE SENTENCE

Europe has started stocking an impressive scientific pantry. There are cloud credits on one shelf. Model access, software licenses and datasets sit on another. A supercomputer waits in the back. Training, internships, research collaborations and technical expertise fill the awkward spaces between them. The cupboard is no longer empty. Now somebody has to explain who gets a key. On October 15, at the European AI in Science Summit in Dublin, the European Commission launched the RAISE AI in Science Enablers Initiative. It asks companies and philanthropic organizations to make voluntary contributions that help European researchers use artificial intelligence in scientific work. RAISE stands for Resource for AI Science in Europe. The broader program is designed as a virtual research institute that pools computing power, data, talent and research funding. The new Enablers initiative adds private and philanthropic resources to that public structure. The first list contains seven organizations: Elsevier, Frontiers, Google, Mistral AI, the Novo Nordisk Foundation, NVIDIA and Sanofi. This is not one giant fund with a single application form. It is a collection of different promises. Google says it is dedicating $3 million in Google Cloud credits and tokens to European researchers and opening direct access to Google DeepMind scientific models. Mistral says it will provide open-weight models, Vibe seats and application programming interface credits, while supporting PhD internships, part-time research roles and university collaborations. The Novo Nordisk Foundation intends to make capacity on Gefion, its AI-focused supercomputer, available to European researchers. NVIDIA is offering openly available scientific software and models, training, technical expertise and pathways to computing and research support. Frontiers says it will provide AI-powered curation for 1,000 Horizon Europe datasets and 100 FAIR-squared data publication grants for early-career researchers, representing more than €1.6 million in services. Elsevier lists complimentary LeapSpace licenses, a Europe-wide challenge, funded research collaborations, AI laboratories, training and access to its content, data and technology. Sanofi points to Future4Care, its health-innovation ecosystem connecting startups, researchers, healthcare organizations and technology companies. That is a real inventory, not a ceremonial bowl of branded pens. It is also an uneven one. A cloud credit behaves differently from a software license. A temporary seat is not the same as an open model. A research collaboration can include human expertise that a self-service portal cannot reproduce. Access to a supercomputer can transform a project, but only if the researcher has code, data, staff and time ready when the allocation arrives. The word pledge is doing important work here. As of October 17, the Commission's public initiative page identifies the first organizations and describes their contributions. It does not yet provide one complete operating manual covering eligibility, application routes, allocation criteria, review panels, geographic distribution, start dates, duration, renewal, usage limits, publication conditions or conflict-of-interest rules across the whole initiative. That does not mean those arrangements do not exist inside individual programs. It means a researcher looking at the shared European initiative cannot yet see them in one place. This is the plain signal: a scientific resource is not truly shared until access can be observed. Europe should publish an access ledger for RAISE. The ledger does not need to expose confidential research proposals or personal data. It should show what was pledged, when it became usable, who administers it, which countries and disciplines can apply, how applicants are chosen, what restrictions follow the resource and what happened after access was granted. For cloud credits, that means reporting the value awarded, the usable services, expiration dates, geographic limits, identity requirements and whether credits cover storage and data movement as well as computation. For model access, it means naming the version, interface, rate limit, retention policy, research-use terms and procedure for preserving results when the model changes. For software and data licenses, the ledger should say whether access survives publication, whether collaborators outside the funded institution can participate and whether a reader can reproduce the method later. For supercomputer allocations, it should record queue time, awarded capacity, actual consumption, technical support and the share of projects that could not use their full allocation. For training and collaborations, it should count more than attendance. Did a participant deploy a workflow? Did a laboratory gain durable capacity? Did an early-career researcher receive authorship, funding or continued access after the workshop ended? The need for this ledger becomes clearer when the public and private parts of RAISE are placed together. The Commission says the RAISE pilot began with €33 million in dedicated Horizon Europe funding in the 2025 work program and €107 million in the 2026 to 2027 work program. It describes five scientific pilot areas: fundamental physics and astronomy, materials science, life science, earth sciences, and social sciences and humanities. Public funding comes with formal calls, budgets and reporting systems. In-kind contributions are usually stranger creatures. Their sticker price can be large while their practical value depends on eligibility, contract terms, staff support, timing and whether the researcher would have purchased the resource anyway. A €10,000 credit that expires before ethics approval is not a €10,000 scientific outcome. A model license that prevents independent auditing can speed up an experiment while making the published result harder to reproduce. A dataset curation service may be more valuable than a much larger compute grant if bad metadata is the actual bottleneck. None of this makes corporate participation suspect by default. Scientific work runs on a mixture of public infrastructure, commercial tools, foundations, universities and individual expertise. Europe needs more pathways for researchers to reach advanced computing and models. Smaller laboratories often cannot negotiate enterprise contracts or maintain specialized AI infrastructure. A coordinated initiative could reduce that disadvantage. But coordination can also concentrate gatekeeping. If every contribution has a separate application, contract and technical environment, well-connected institutions will be better equipped to navigate the pantry. If allocation rules reward projects that are already polished, researchers who most need help building capability may arrive too early to qualify. If donors select projects independently, the scientific agenda can drift toward fields that fit their products or reputations. The Commission's charter commits the initiative to scientific excellence, responsible and trustworthy AI, collaboration, and applicable legal and ethical standards. Those are sensible principles. They are not allocation rules. An access ledger would make the principles testable. It could show whether resources reach researchers across member states rather than clustering around a few famous institutions. It could separate access announced from access activated, and activated from access used. It could record disciplinary balance without pretending that every field requires equal amounts of computation. It could also disclose recusals and donor relationships when an organization helps choose the projects that receive its own product. Reproducibility needs its own column. Science is not finished when a model produces a promising result. Other researchers need enough information and access to test it. Commercial models change. Application programming interfaces are retired. Credit programs end. Licensed datasets can remain unavailable to reviewers. A project can be methodologically sound on publication day and practically impossible to repeat a year later. RAISE can address that by requiring each supported project to publish a resource receipt alongside its paper or report. The receipt should identify model and software versions, compute consumed, data access conditions, material donor restrictions, conflicts of interest and a realistic reproduction route. Where the original resource cannot remain available, the project should say so plainly. A smaller open benchmark, archived output or independent replication grant may provide a partial bridge. The answer cannot be to hide the dependency behind the phrase AI-enabled research. The location of the launch matters too. Dublin is a European technology center and the summit brought researchers, companies, investors and policymakers together. Hosting a pledge ceremony in Ireland does not tell us whether an Irish university outside the largest networks, a laboratory in Romania, a climate team in Portugal or a humanities group in Croatia can obtain the same practical access. Only allocation data can answer that. The most encouraging detail in the Commission's new page is that it names concrete contributions instead of merely inviting vague partnership. There are amounts, services, machines and tools to examine. That gives the initiative something solid enough to audit. The next move should be equally concrete. Publish the inventory. Publish the application doors. Publish the decision rules. Publish the delays, denials, expirations and actual use. Connect every scientific result back to the resources that made it possible. Europe has assembled the pantry. Now it needs to show that the key does not belong only to the people already standing behind the counter.