AbstractWhen simulating past warm climates, such as the early Cretaceous and Paleogene periods, general circulation models (GCMs) underestimate the magnitude of warming in the Arctic. Additionally, model intercomparisons show a large spread in the magnitude of Arctic warming for these warmer-than-modern climates. Several mechanisms have been proposed to explain these disagreements, including the unrealistic representation of polar clouds or underestimated poleward heat transport in the models. This study provides an intercomparison of Arctic cloud and atmospheric heat transport (AHT) responses to strong imposed polar-amplified surface ocean warming across four atmosphere-only GCMs. All models simulate an increase in high clouds throughout the year; the resulting reduction in longwave radiation loss to space acts to support the imposed Arctic warming. The response of low and mid-level clouds varies considerably across the models, with models responding differently to surface warming and sea ice removal. The AHT is consistently weaker in the imposed warming experiments due to a large reduction in dry static energy transport that offsets a smaller increase in latent heat transport, thereby opposing the imposed surface warming. Our idealised polar amplification experiments require very large increases in implied ocean heat transport (OHT) to maintain steady state. Increased CO2 or tropical temperatures that likely characterised past warm climates, reduces the need for such large OHT increases.