EU AI science plan: big goals, tight clock - coordination will decide the outcome
The European Commission set a high bar on 8 October with a plan to boost AI in science and speed industry adoption. At the center is a virtual institute, the Resource for AI Science in Europe (Raise), meant to connect talent, computing, data, and funding across the bloc.
The ambition is welcome. The risk is timing. Doubling AI research funding by 2028 sounds strong on paper, but leaders warn that without fast coordination between the EU, member states, and businesses, the impact may arrive after the critical window.
What Raise must deliver fast
- Simple, fair access to large-scale compute (linked to existing capacity such as EuroHPC JU).
- Federated, high-quality data commons with shared governance, provenance, and consent tooling.
- Common evaluation protocols and reproducibility standards for models and datasets.
- Mobility pathways for researchers across universities, RIs, and industry labs.
- Streamlined funding instruments that support cross-border, multi-institution consortia.
- Clear IP, licensing, and procurement rules to speed tech transfer and public adoption.
The coordination gap
Europe's research fabric is strong but fragmented. National strategies, university priorities, and corporate roadmaps often move in parallel rather than together.
- Compute allocation differs by country and program - leading to waitlists and underuse elsewhere.
- Duplicate investments in similar model pipelines add cost without improving outcomes.
- Data access is uneven; legal uncertainty slows sharing even when it is permissible.
- Salary and hiring rules make it hard to recruit or retain ML engineers and research software engineers.
- Cross-border grants take time, creating a lag that hurts iteration speed.
Funding: doubling by 2028 is not enough by itself
More money helps, but back-loaded budgets won't clear 2025-2026 bottlenecks. The priority is front-loading capacity that researchers can actually use.
- Bring forward compute procurement and sustain the opex for training and inference, not just capex.
- Fund shared tooling: evaluation harnesses, MLOps, data curation pipelines, and safety testing.
- Design calls that require multisite collaboration and shared infrastructure from day one.
- Back challenge-led programs in climate, health, materials, energy, and AI safety to focus effort.
Policy alignment that reduces friction
Regulatory clarity increases confidence for interdisciplinary projects and clinical or industrial deployments. Align internal processes early with EU rules to avoid last-minute rework.
- Map compliance plans to the EU's approach to AI and existing data protection rules such as GDPR. See the Commission's overview of AI policy here.
- Use standardized model documentation, risk classification, and incident reporting across projects.
Practical steps research leaders can take now
- Build a shared compute plan across your institution and national partners; pre-book capacity where possible.
- Stand up a data governance board to approve datasets, consent models, and cross-border sharing.
- Adopt common evaluation suites and publish benchmark results for all funded model work.
- Create joint appointments with industry labs; negotiate IP and publishing terms upfront.
- Fund research software engineering teams as first-class partners, not support afterthoughts.
- Run quarterly grant sprints that match EU calls with internal proposals and external collaborators.
Risks to watch
- Compute scarcity, long queues, or fragmented access policies.
- IP and licensing disputes that stall pilots and scale-up.
- Security constraints and export controls that limit model sharing.
- Ethics reviews that start late and delay trials or deployments.
- Vendor lock-in that raises costs and blocks portability.
What good looks like by 2028
- Time-to-compute measured in days, not months, for cross-border teams.
- Federated datasets actively used in high-impact projects across multiple member states.
- Transparent, reproducible benchmarks and evaluation reports for all major models.
- Clinical- or industry-grade AI systems deployed with clear safety and monitoring plans.
- Career paths that keep top ML and data talent inside European research and industry.
Upskilling your team
Skills will limit throughput as much as hardware. If your lab or center is scaling AI work, align training with project needs and roles.
- Explore role-based AI course paths for research teams: Courses by job.
- For structured credentials to support hiring and promotion, see AI certification options.
The EU's plan can lift scientific output - if coordination closes the gap between policy and lab benches. Move early on compute, data, evaluation, and talent, and the funding will compound rather than catch up.
Your membership also unlocks:
