Ice shelves fringing the Antarctic coastline limit sea-level rise by slowing down the flow of grounded ice into the sea. They are thinning because of warming and intensifying ocean currents. Building upon my expertise in multi-scale flow dynamics and subglacial oceanography, I will resolve the two main bottlenecks that impair our ability to assess and project the state of Antarctic ice shelves. These are: (i) a poor understanding of the relationship between ice-shelf ablation rates and ocean properties & (ii) the high computational cost of ocean simulations. I will first build new models of the metre-thick ice-shelf—ocean boundary layer (ISOBL), using innovative laboratory experiments and simulations that will unravel its dynamics. The lack of accurate ISOBL models is responsible for leading-order errors in predicting ice loss rates and freshwater production. The new models will capture how turbulence controls heat fluxes over an unprecedented range of sub ice-shelf conditions, accurately predicting ice ablation rates from the grounding line to the shelf front. I will then promote a paradigm shift in the modelling of oceans beneath ice shelves, which are hundreds of kilometres wide. At present, ensemble simulations of ice-shelf ocean cavities on 100-year time scales are about 1000 times computationally too costly to resolve mesoscales, which influence the mean circulation and ice ablation. To circumvent this issue, I will train novel data-driven reduced-order models (ROMs), which will emulate the ocean dynamics at high resolution and unprecedently low computational cost. The ROMs will learn the key fingerprints of ocean cavities and their time evolution from short-term high-resolution simulations. They will then be extended to longer times and different forcing conditions, enabling eddy-resolving IPCC-level large ensemble simulations of subglacial oceans, which will help reduce uncertainty related to the timing of abrupt regime changes of the Antarctic ice sheet.