ECMWF's Artificial Intelligence Forecasting System (AIFS) is built on a graph neural network encoder-processor-decoder architecture that converts gridded atmospheric data into polyhedral graph structures for efficient message passing [1][2]. The single-model configuration progressed from experimental version v0.2.1 to the first operational release v1.0 in February 2025, followed by v1.1.0, which introduced physical consistency constraints, expanded forecast variables, and optimized training protocols at approximately 0.25-degree resolution [3][2]. The ensemble configuration extends this same AI framework to probabilistic forecasting by generating multiple members with stochastic perturbations to quantify atmospheric uncertainty across a 15-day forecast horizon [2]. Both configurations share the core AI architecture and training data from ERA5 reanalysis, but the ensemble version specifically captures forecast spread through member diversity. AIFS represents a transition from research prototype to operational tool, demonstrating computational efficiency and forecast skill improvements over traditional numerical weather prediction models [1][2].