Permafrost thaw is transforming Arctic landscapes, yet why thaw progresses differently across space and time remains poorly understood. Although ice-wedge degradation is widespread across the Arctic, its spatial organization and temporal evolution have not been well constrained, limiting our ability to forecast how the Arctic will change as it continues warming. Using a decadal time series of sub-meter satellite imagery and a U-Net deep learning classifier, we reconstructed fine-scale trajectories of ice-wedge degradation across a 50-km² Arctic landscape. Ice-wedge degradation did not follow a single progression toward advanced thermokarst. Instead, thaw evolved through multiple asynchronous and non-linear trajectories that were organized by geomorphic landforms across spatial scales. The location, timing, and severity of degradation varied systematically among landforms, indicating that millennia-scale geomorphic history sets the template for modern permafrost degradation trajectories. These spatial patterns suggest that climate warming is restructuring rather than homogenizing Arctic permafrost landscapes through interactions between inherited geomorphic setting and local ecohydrological feedbacks. Our framework of considering the spatial and temporal organization of permafrost thaw, not simply its extent or severity, provides a new basis for predicting Arctic landscape evolution and its hydrologic, ecological, and biogeochemical responses to continued warming.