Verily Health, collaborating with NVIDIA and the NIH All of Us Research Program, built an early multimodal foundation model that integrates electronic health record and genomic data. The work provides a concrete example of how multimodal foundation models can combine EHR “lived experience” with inherited risk signals from genomics. The model is intended to move beyond single-modality learning, reflecting the view that EHRs represent only part of a patient’s health history and that integrating additional modalities can improve disease prediction and personalization. The effort also highlights practical challenges in assembling and aligning multimodal datasets—an issue likely to shape how healthcare AI systems are trained and deployed.