A deep learning model trained on 26,134 UK Biobank electrocardiograms can predict left atrial imaging markers and diagnose atrial cardiomyopathy—a condition that precedes atrial fibrillation, heart failure, and stroke. The AI system outperformed established ECG markers and significantly improved risk stratification beyond current clinical scoring systems like CHARGE-AF. This breakthrough matters because atrial cardiomyopathy represents a critical early stage of heart disease that's traditionally diagnosed through expensive imaging tests. By extracting this information from routine 12-lead ECGs—available in virtually every healthcare setting—the technology could enable widespread early detection of cardiovascular risk. The model's ability to predict heart failure risk even in patients without atrial fibrillation suggests it captures fundamental structural heart changes that precede clinical symptoms. However, this preprint awaits peer review, and validation beyond the UK Biobank population will be essential. While promising for screening applications, the technology requires careful clinical integration to avoid false positives in low-risk populations. This represents an incremental but potentially impactful advance in cardiovascular risk prediction, democratizing access to sophisticated cardiac assessment through ubiquitous ECG technology.