Fast pairwise grouping methods have recently shown great promising for large-scale clustering. However, the relationships among objects of the real world are often high-order rather than pairwise. Naively applying pairwise grouping methods to high-order grouping tasks will not always obtain the desired clustering results since there is a loss of information. In this paper, we propose a fast hypergraph clustering algorithm via the Nyström Extension. As the hyperedge of the hypergraph allows us to join any number of vertices, high-order relationships among objects can be captured. In contrast to some previous works in the Nyström research focusing on the error bound for the Nyström-approximated kernel matrix, we present an error bound for the approximated eigenvectors associated with the Nyström Extension of hypergraph Laplacian. In addition, a novel formulation for sampling size selection is provided. Our experiments on a series of UCI data sets show that the proposed algorithm achieves significantly speeding and lower storage. Specifically, our algorithm outperforms the pairwise spectral grouping method based on the Nyström method in terms of accuracy, computation time and space occupation.