EGI Federation reviewed its early AI work in interTwin and iMagine projects. It noted a shift in science AI from model training to domain-specific adaptation. EGI presented a new AI strategy for 2026-2030 built on three enablers. The first enabler is an AI-ready infrastructure linking compute, data and applications. The second is human skills for AI implementation. The third is trust for high-risk AI environments. EGI described projects for federated LLM inference, AI-ready data platforms and competence centres. It outlined a division of labour with EuroHPC, RAISE, research infrastructures and EOSC. EGI will specialise in customising and operating AI for research long term.