SeaCast, an AI-powered ocean forecasting system, delivers highly detailed 15-day predictions for the Mediterranean Sea in just seconds. Developed by researchers from the CMCC Foundation and the University of Helsinki, the model uses a graph neural network (GNN) to perform autoregressive ocean forecasting, enabling it to generate extended forecasts through repeated cycles of encoding, processing, and decoding. Unlike most global AI models that operate at lower resolutions and rely solely on ocean data, SeaCast integrates both oceanic and atmospheric variables. This fusion of data allows the model to capture complex regional dynamics, particularly near coastlines and at boundaries, where traditional models often struggle. The system operates at a high resolution of about 4 km (1/24°), matching the detail of the CMCC’s operational Mediterranean forecasting system MedFS, which is used by the Copernicus Marine Service. Training on CMCC Mediterranean reanalysis data—freely available through Copernicus—the model produces forecasts up to 200 meters deep. Its performance surpasses the current operational numerical model over a standard 10-day forecast window and extends predictions to 15 days, a significant improvement in lead time. The efficiency gains are dramatic. While the traditional numerical system takes about 70 minutes using 89 CPUs to generate a 10-day forecast, SeaCast completes a 15-day forecast in just 20 seconds using a single GPU. This leap in speed and energy efficiency makes it possible to run rapid scenario testing and probabilistic ensemble forecasts—critical tools for assessing uncertainty in ocean predictions. These capabilities have real-world applications in shipping, aquaculture, environmental monitoring, and coastal risk management. Fast, accurate forecasts enable better planning and proactive responses to changing sea conditions. A key innovation is SeaCast’s integration of atmospheric forcing data during training and forecasting. Sensitivity experiments show that including atmospheric variables significantly improves accuracy, especially near the ocean surface where atmospheric interactions are strongest. Longer training periods—up to 35 years of historical data—further enhance model performance. The project highlights the power of interdisciplinary collaboration. Researchers combined expertise in oceanography, atmospheric science, and AI to overcome challenges that neither field could solve alone. As Emanuela Clementi, a CMCC researcher and co-author, noted, this synergy enables more accurate forecasts at a fraction of the computational cost. Daniel Holmberg, the study’s first author, emphasized the value of working closely with CMCC, including field visits and collaborative discussions that enriched the research process. Looking ahead, CMCC plans to integrate SeaCast into operational forecasting systems, complementing traditional physics-based models. As the first high-resolution, regional AI ocean model, SeaCast sets a new benchmark and paves the way for smarter, faster, and more reliable marine predictions worldwide.