Neural Kernel Surface Reconstruction (NKSR) is a CVPR 2023 Highlight paper from NVIDIA that reconstructs 3D implicit surfaces from large-scale point clouds. The method builds on Neural Kernel Fields and uses compactly supported kernel functions with sparse linear solvers to scale to large scenes, while being robust to noise through gradient fitting. It achieves state-of-the-art results on reconstruction benchmarks for single objects, indoor scenes, and outdoor scenes, handling millions of points in seconds.