CARDIAC-FM, a multimodal AI foundation model, demonstrated superior cardiovascular risk prediction by learning from 57,609 paired ECG and cardiac MRI samples from UK Biobank. The model outperformed single-modality approaches across external validation cohorts including the Cardiovascular Health Study and Multi-Ethnic Study of Atherosclerosis, showing particular promise for predicting atrial fibrillation and heart failure outcomes. The foundation model approach represents a significant advancement in cardiovascular AI, as it can generalize across multiple heart conditions with minimal retraining—addressing a key limitation where rare diseases lack sufficient data for dedicated prediction models. Importantly, while trained on both ECG and MRI data, the system can generate clinical predictions using only ECG input, making it practically deployable in standard clinical settings without expensive imaging requirements. This accessibility could democratize advanced cardiovascular risk assessment. However, this preprint awaits peer review, and the real-world clinical validation remains to be established. The approach appears confirmatory of multimodal AI's superiority while offering practical deployment advantages that could meaningfully impact cardiovascular care if validated through rigorous clinical trials.