As part of DOE's ECP Center for Efficient Exascale Discretizations, we have been developing Nek5000, NekCEM, and a new, GPU-oriented code, NekRS, for a variety of scientific applications on DOE's leadership computing facilities. The primary focus is on high-fidelity turbulent flow and heat transfer applications, for example, as used extensively in the reactor-design area. Other applications include aerosol transport for Covid-19, internal combustion engines, wind-farm modeling, transport of contaminants in the ocean, vascular flow simulation, residential HVAC, ion transport, and fundamental studies in turbulence, combustion, fluid-structure interaction, and electromagnetics. In Part 1 of this talk, we describe the challenges of parallelism at exascale and how we have addressed these challenges in the development of NekRS, which readily strong-scales to all of Summit (27648 NVIDIA V100 GPUs). In Part 2, we present extensive scaling analysis for reactor thermo-hydraulics and atmospheric boundary layer simulations. We will also discuss some ongoing works on particle tracking and molten salt reactor simulations.