The basic science responsible for the fascinating shapes of ice crystals and snowflakes is still not understood. Insufficient knowledge of the interaction potentials and the lack of relevant experimental access to the growth process are to blame for this failure. Here, we study the growth of fractal nanostructures in a two-dimensional (2D) system, intercalated between mica and graphene. Based on our scanning tunneling spectroscopy data, we provide compelling evidence that these fractals are 2D ice. They grow while they are in material contact with the atmosphere at 20°C and without significant thermal contact to the ambient. The growth is studied in situ, in real time and space at the nanoscale. We find that the growing 2D ice nanocrystals assume a fractal shape, which is conventionally attributed to Diffusion Limited Aggregation (DLA). However, DLA requires a low mass density mother phase, in contrast to the actual currently present high mass density mother phase. Latent heat effects and consequent transport of heat and molecules are found to be key ingredients for understanding the evolution of the snow (ice) flakes. We conclude that not the local availability of water molecules (DLA), but rather them having the locally required orientation is the key factor for incorporation into the 2D ice nanocrystal. In combination with the transport of latent heat, we attribute the evolution of fractal 2D ice nanocrystals to local temperature dependent rotation limited aggregation. The ice growth occurs under extreme supersaturation, i.e., the conditions closely resemble the natural ones for the growth of complex 2D snow (ice) flakes and we consider our findings crucial for solving the "perennial" snow (ice) flake enigma.